No full-text available
To read the full-text of this research,
you can request a copy directly from the authors.
Ensemble Meta Learning for Few-Shot Soot Density Recognition
Abstract
In each petrochemical plant around the world, the flare stack as a requisite facility produces a large amount of soot due to the incomplete combustion of flare gas, and this strongly endangers air quality and human health. Despite severe damage, the above-mentioned abnormal conditions rarely occur and thus only few-shot samples are available. To address such difficulty, we design an image-based flare soot density recognition network (FSDR-Net) via a new ensemble meta-learning technology. More particularly, we first train a deep convolutional neural network (CNN) by applying the model-agnostic meta-learning algorithm on a variety of learning tasks that are relevant to the flare soot recognition, so as to obtain the general-purpose optimized initial parameters (GOIP). Second, for the new task of recognizing the flare soot density via only few-shot instances, a new ensemble is developed to selectively aggregate several predictions which are generated based on a wide range of learning rates and a small number of gradient steps. Results of experiments conducted on the density recognition of flare soot corroborate the superiority of our proposed FSDR-Net as compared with the popular and state-of-the-art deep CNNs.
... After image enhancement and segmentation, feature design and extraction are the third phases of a medical detection system. Some features are designed by an expert (e.g., shape, size, and brightness), some are designed by algorithms (e.g., clustering algorithm), and some use a combination of both [14][15][16][17]. In Pap smear cytology images, up to 20 features are designed by an expert for the classification phase. ...
... William et al. [35] employed Fuzzy C-means to classify the images, but a classical approach was used for image enhancement, segmentation, and feature extraction. Also, Mohammad Subhi et al. [14], for uncertainty modelling, investigated an automatic feature extraction (AFE) algorithm with a neurofuzzy model called multiple adaptive neuro-fuzzy inference system (MANFIS) for the recognition process. The results were promising but on a local dataset. ...
Applications of deep learning models for medical image analysis have been concentrated in recent years. This study presents an improved automatic Deep Learning Ensemble model to detect cervical cancer. The proposed model takes advantage of the Deep Belief Network and Ensemble classification model called EDBM to improve the accuracy of the prediction. In this method, a new classification layer using Belief Network with Dempster combinational rule combines the evidence to handle uncertainty. An ensemble classifier was applied to address the issue of sample rejection in previous classification models. This method used three classifiers, including Support Vector Machine, K-Nearest Neighbour, and Fuzzy classifier, to improve the model's performance. Experimental results on the Herlev dataset and the SIPaKMeD dataset revealed the superiority of the proposed EDBM model. The EDBM On the Herlev dataset with an accuracy of 99%, specificity of 98%, the sensitivity of 98%, and AUC of 99.98%, and the SIPaKMeD dataset with an accuracy of 97.2%, specificity of 98.79%, the sensitivity of 98.70%, and AUC of 99.89%, outperformed counterpart Deep learning methods and showed promising achievements in the early detection of Cervical Cancer.
... Along with the ever-growing number of connected Internet of Things (IoT) devices, machine learning (ML) has been Chi-Hieu Nguyen, D. T. Hoang, D. N. Nguyen playing a critical role in learning and extracting knowledge from distributed data sources, e.g., from digital healthcare [1]- [3], Industry 4.0 [4]- [6], and the emerging Metaverse [7], [8]. Among various distributed learning frameworks, federated learning (FL) has been considered as the most potential one for the mobile edge computing (MEC) networks to reduce communication overhead and address the users' privacy issues [9], [10]. ...
... Likewise, the encrypted training process carried out at an MEN-k is achieved using its processor's frequency f k (Hz) which requires η m (cycles/bit) for 1-bit data training. Since the total encrypted dataset collected at MUs has size of N n=1 x n,k b n (bits), the computation time of an MEN-k can be obtained by 6 Regarding the communication model, let sΓ(r), s Γ (r) n , and s Γ (r) k denote the sizes in bits of the global model, the local trained model at MU-n, and the MEN-aggregated model at MEN-k, respectively, in which sΓ(r) = s Γ (r) n = s Γ (r) k [15], [41]. The number of successful transmissions for the uplink and downlink communication between the CS, MEN-k, and MU-n are, respectively, defined as ν up k,K , ν up n,K , ν up n,k , and ν down k,K , ν down n,K , ν down n,k . ...
p>Federated Learning (FL) plays a pivotal role in enabling artificial intelligence (AI)-based mobile applications in mobile edge computing (MEC). However, due to the resource heterogeneity among participating mobile users (MUs), delayed updates from slow MUs may deteriorate the learning speed of the MEC-based FL system, commonly referred to as the straggling problem. To tackle the problem, this work proposes a novel privacy-preserving FL framework that utilizes homomorphic encryption (HE) based solutions to enable MUs, particularly resource-constrained MUs, to securely offload part of their training tasks to the cloud server (CS) and mobile edge nodes (MENs). Our framework first develops an efficient method for packing batches of training data into HE ciphertexts to reduce the complexity of HE-encrypted training at the MENs/CS. On that basis, the mobile service provider (MSP) can incentivize straggling MUs to encrypt part of their local datasets that are uploaded to certain MENs or the CS for caching and remote training. However, caching a large amount of encrypted data at the MENs and CS for FL may not only overburden those nodes but also incur a prohibitive cost of remote training, which ultimately reduces the MSP’s overall profit. To optimize the portion of MUs’ data to be encrypted, cached, and trained at the MENs/CS, we formulate an MSP’s profit maximization problem, considering all MUs’ and MENs’ resource capabilities and data handling costs (including encryption, caching, and training) as well as the MSP’s incentive budget. We then show that the problem is convex and can be efficiently solved using an interior point method. Extensive simulations on a real-world human activity recognition dataset show that our proposed framework can achieve much higher model accuracy (improving up to 24.29%) and faster convergence rate (by 2.86 times) than those of the conventional FedAvg approach when the straggling probability varies between 20% and 80%. Moreover, the proposed framework can improve the MSP’s profit up to 2.84 times compared with other baseline FL approaches without MEN-assisted training.</p
... Along with the ever-growing number of connected Internet of Things (IoT) devices, machine learning (ML) has been Chi-Hieu Nguyen, D. T. Hoang, D. N. Nguyen playing a critical role in learning and extracting knowledge from distributed data sources, e.g., from digital healthcare [1]- [3], Industry 4.0 [4]- [6], and the emerging Metaverse [7], [8]. Among various distributed learning frameworks, federated learning (FL) has been considered as the most potential one for the mobile edge computing (MEC) networks to reduce communication overhead and address the users' privacy issues [9], [10]. ...
... Likewise, the encrypted training process carried out at an MEN-k is achieved using its processor's frequency f k (Hz) which requires η m (cycles/bit) for 1-bit data training. Since the total encrypted dataset collected at MUs has size of N n=1 x n,k b n (bits), the computation time of an MEN-k can be obtained by 6 Regarding the communication model, let sΓ(r), s Γ (r) n , and s Γ (r) k denote the sizes in bits of the global model, the local trained model at MU-n, and the MEN-aggregated model at MEN-k, respectively, in which sΓ(r) = s Γ (r) n = s Γ (r) k [15], [41]. The number of successful transmissions for the uplink and downlink communication between the CS, MEN-k, and MU-n are, respectively, defined as ν up k,K , ν up n,K , ν up n,k , and ν down k,K , ν down n,K , ν down n,k . ...
p>Federated Learning (FL) plays a pivotal role in enabling artificial intelligence (AI)-based mobile applications in mobile edge computing (MEC). However, due to the resource heterogeneity among participating mobile users (MUs), delayed updates from slow MUs may deteriorate the learning speed of the MEC-based FL system, commonly referred to as the straggling problem. To tackle the problem, this work proposes a novel privacy-preserving FL framework that utilizes homomorphic encryption (HE) based solutions to enable MUs, particularly resource-constrained MUs, to securely offload part of their training tasks to the cloud server (CS) and mobile edge nodes (MENs). Our framework first develops an efficient method for packing batches of training data into HE ciphertexts to reduce the complexity of HE-encrypted training at the MENs/CS. On that basis, the mobile service provider (MSP) can incentivize straggling MUs to encrypt part of their local datasets that are uploaded to certain MENs or the CS for caching and remote training. However, caching a large amount of encrypted data at the MENs and CS for FL may not only overburden those nodes but also incur a prohibitive cost of remote training, which ultimately reduces the MSP’s overall profit. To optimize the portion of MUs’ data to be encrypted, cached, and trained at the MENs/CS, we formulate an MSP’s profit maximization problem, considering all MUs’ and MENs’ resource capabilities and data handling costs (including encryption, caching, and training) as well as the MSP’s incentive budget. We then show that the problem is convex and can be efficiently solved using an interior point method. Extensive simulations on a real-world human activity recognition dataset show that our proposed framework can achieve much higher model accuracy (improving up to 24.29%) and faster convergence rate (by 2.86 times) than those of the conventional FedAvg approach when the straggling probability varies between 20% and 80%. Moreover, the proposed framework can improve the MSP’s profit up to 2.84 times compared with other baseline FL approaches without MEN-assisted training.</p
... Meta learning is defined as the learning algorithm which learn from other learning algorithms. This means that the learning model already learn how to integrate the predictions generated from the base model to a single model [63]. Single machine learning model suffer from various challenges like lack of high-quality training datasets and high time complexity of the algorithms to generate predictions. ...
... This SSEP model is superior to state -of -theart competitors and intermediate models. An image-based flare soot density recognition network (FSDR-Net) is designed by the authors of [24]. FSDR-Net is trained by using deep convolutional network with the soot images that were taken in the real petrochemical plant which gives the high performance more than the modern deep networks. ...
Many researchers have preferred non-invasive techniques for recognizing the exact type of physiological abnormality in the vocal tract by training machine learning algorithms with feature descriptors extracted from the voice signal. However, until now, most techniques have been limited to classifying whether a voice is normal or abnormal. It is crucial that the trained Artificial Intelligence (AI) be able to identify the exact pathology associated with voice for implementation in a realistic environment. Another issue is the need to suppress the ambient noise that could be mixed up with the spectra of the voice. Current work proposes a robust, less time-consuming and non-invasive technique for the identification of pathology associated with a laryngeal voice signal. More specifically, a two-stage signal filtering approach that encompasses a score-based geometric approach and a glottal inverse filtering method is applied to the input voice signal. The aim here is to estimate the noise spectra, to regenerate a clean signal and finally to deliver a completely fundamental glottal flow-derived signal. For the next stage, clean glottal derivative signals are used in the formation of a novel fused-scalogram which is currently referred to as the "Combinatorial Transformative Scalogram (CTS)." The CTS is a time-frequency domain plot which is a combination of two time-frequency scalograms. There is a thorough investigation of the performance of the two individual scalograms as well as that of the CTS database.Nine classification metrics are used to investigate performance, which are: sensitivity, mean accuracy, error, precision, false positive rate, specificity, Cohen’s kappa, Matthews Correlation Coefficient, and F1 score. Implementation of the VOice ICar fEDerico II (VOICED) standard database provided the highest mean accuracy of 94.12% with a sensitivity of 93.85% and a specificity of 97.96% against other existing techniques. The current method performed well despite the data imbalance that exists between classes.
... Based on the above reasons, only using a small amount of labeled data to achieve object detection has become a popular topic in the field of deep learning. Many current few-shot detection methods utilize meta-learning [2,[16][17][18][19] to generalize object detection networks, allowing them to acquire prior knowledge during training and transfer that knowledge to new class tasks. However, meta-learning can be challenging to design and the training process can be complex. ...
In the domain of monitoring and conservation of rare and endangered wildlife, traditional object detection methods fail to operate effectively due to overfitting caused by limited image data. To address this issue, this paper proposes a small-sample object detection method that utilizes limited support information to achieve accurate detection of endangered wildlife images. Firstly, we introduce the Feature Pyramid Network (FPN) to adapt to foreground targets at various scales. Secondly, we propose a Coupled Relationship Activation (CRA) module that aggregates features from the support set into a comprehensive feature vector and utilizes it to activate similar regions in the query feature. Furthermore, to balance the differences between support and query features, we incorporate cosine similarity and short-circuit structures before and after the activation process, thereby mitigating the impact of activating dissimilar features erroneously. During the above process, we construct a balanced dataset, CRBAD, which comprises images of both endangered and common animal categories, and employ it to conduct practical testing of the proposed method. Experimental results demonstrate that the proposed model achieves state-of-the-art performance in detecting endangered wildlife images.
... Gu et al. [131] proposed a deep learning model that combines model-agnostic meta-learning (MAML) algorithm combined with selective ensemble for the density prediction of flare soot using only few-shot data. As the problem of flare soot occurs rarely on very limited data are available for such condition, the use of few-shot learning is of extreme importance. ...
The real-time analysis of gas flares is one of the most challenging problems in the operation of various combustion-involving industries, such as oil and gas refineries. Despite the crucial role of gas flares in securing safe plant operation and lowering environmental pollution, they are among the least monitored components of petrochemical plants due to their harsh working environments. Remote sensing techniques are emerging as potential alternatives for conventional sampling-based techniques in visual inspection and performance analysis. This paper presents an in-depth review of significant achievements in Gas flares monitoring over past two decades and highlights ongoing challenges in Artificial Intelligence (AI)-enhanced remote inspection. By reading the content, both industry professionals and academic researchers can gain a comprehensive knowledge of improvements from the integration of AI and remote sensing, understanding the trend of Gas flares monitoring. The paper commences with an analysis of RGB camera-based methods, focusing on how their combination with cutting-edge machine learning and deep learning algorithms can significantly improve the detection, segmentation, and measurement of gas flare systems. It then explores the use of hyper-spectral imaging techniques, including infrared cameras and space-borne satellite sensors, underscoring their potential in remote monitoring and performance analysis. Additionally, the effectiveness of multi-view inspection methods is assessed, highlighting how these approaches enhance the monitoring capabilities. Finally, the paper identifies key research areas that require further attention. It also presents a clearer direction for future progress, emphasizing the importance of continuous research to foster advancements and facilitate broader commercial adoption.
... In order to solve this problem, deep learning technology has been widely used in the field of industrial design. Deep learning model can automatically generate sketches by learning a large number of design data and patterns [2][3] . This not only provides designers with faster and more efficient design tools, but also provides new possibilities for innovation in the design process. ...
This study is devoted to exploring the strategy of automatic sketch generation and optimization of industrial design based on deep learning. By combining the Generative Adversarial Network (GAN) with the optimization algorithm, this paper proposes an innovative method to realize the automatic generation of high-quality and diverse industrial design sketches. In the experiment, this paper selects SketchyCAD and other public data sets, trains them through deep learning model, and introduces genetic algorithm(GA) and differential evolution algorithm to optimize the parameters. In terms of experimental results, we observed that the quality of generated sketches was significantly improved, and the design sketches generated by the mode (GAN+GA) were more realistic and innovative. The introduction of optimization strategy further improves the generation effect and intelligently adjusts the model parameters to adapt to different design styles. In this paper, the influence of hyperparameter tuning is analyzed in detail, and it is found that the adjustment of learning rate plays a key role in generating quality and diversity. However, the experiment also revealed some challenges and room for improvement. We noticed that the generated results may have the risk of over-fitting in the training process, and with the increase of training times, the diversity gradually decreased. This suggests that more complex model structure and richer data sets are needed to improve the generalization performance. Generally speaking, this study provides new ideas and methods for the integration of deep learning and industrial design. By innovatively combining generation model and optimization algorithm, this research has contributed beneficial research results to the development of industrial design automation. This research is of great significance to promote the intelligence and innovation in the field of industrial design.
... Nowadays, deep learning has been successfully applied to a variety of applications (Song et al., 2016;Qiu et al., 2018;Zhang SQ et al., 2018;Gu et al., 2020Gu et al., , 2021aGu et al., , 2021b and achieved remarkable success. Kang et al. (2014) were the first to apply a CNN to the NR-IQA. ...
Technological advancements continue to expand the communications industry’s potential. Images, which are an important component in strengthening communication, are widely available. Therefore, image quality assessment (IQA) is critical in improving content delivered to end users. Convolutional neural networks (CNNs) used in IQA face two common challenges. One issue is that these methods fail to provide the best representation of the image. The other issue is that the models have a large number of parameters, which easily leads to overfitting. To address these issues, the dense convolution network (DSC-Net), a deep learning model with fewer parameters, is proposed for no-reference image quality assessment (NR-IQA). Moreover, it is obvious that the use of multimodal data for deep learning has improved the performance of applications. As a result, multimodal dense convolution network (MDSC-Net) fuses the texture features extracted using the gray-level co-occurrence matrix (GLCM) method and spatial features extracted using DSC-Net and predicts the image quality. The performance of the proposed framework on the benchmark synthetic datasets LIVE, TID2013, and KADID-10k demonstrates that the MDSC-Net approach achieves good performance over state-of-the-art methods for the NR-IQA task.
... The energy function plays a major role in eliminating the manual operations, however the method was applied on a small dataset. Gu et al. developed segmentation models for soot recognition [21], smoke detection [22] and P M 2.5 monitoring [23] from images using ensemble meta-learning, dual deep channel neural networks and incorporation of information abundance measurement using deep learning models respectively. Density based clustering [24] has been explored for segmenting medical images effectively. ...
Delineation of Gallbladder (GB) and identification of gallstones from Computed Tomography (CT) and Ultrasonography (USG) images is an essential step in the radiomic analysis of Gallbladder Cancer (GBC). In this study, we devise a method for effective segmentation of GB from 2D CT images and Gallstones from USG images, by introducing a Rough Density based Segmentation (RDS) method. Based on the threshold value obtained using rough entropy thresholding, the image is thresholded and passed as an input to the RDS method to obtain the desired segmented regions. To evaluate the performance of RDS method, we collected images from 30 patients exhibiting normal GB and 8 patients with gallstones. Additionally, the versatility of our RDS method has also been tested for segmenting lungs from a publicly available Covid-19 lung CT image dataset with cohort size of 20 patients. Our method has been compared with several well-known methods like hybrid fuzzy clustering, morphological active contour without edges, modified fuzzy c means and morphological geodesic active contours and found to give significantly better results with reference to Jaccard coefficient, Dice coefficient, accuracy, precision, sensitivity, specificity and McNemar’s test.
... Authors in [25] treated the few-shot learning problem in the recognition of flare soot density with the meta-learning technics which is used to extract the intrinsic features from a bunch of similar tasks (e.g., dark clouds, cigarette smoke, and vapor). Then the obtained weights are used as initial values to train the network in the main task i.e. soot density recognition. ...
Deep Learning has significantly push forward the research on cancer magnetic resonance imaging (MRI) images. These images are widely used to diagnose the presence of a deformed tissue within the brain in which the cells replicate indefinitely without control, i.e. a brain tumor. Radiologist have to deeply examine a set of MRI images for each patient in order to decide whether the tumor is benign (noncancerous) or malignant (cancerous). The latest have very severe consequences and have a very high mortality rate, but this could be significantly reduced if the cancer is diagnosed at an earlier stage. The classification task is very complicated due to neurological and radiological similarities of different tumors. In order to assist the radiologists, our objective in this paper is to achieve a correct classification of the MRI images. The studied deep classification models have been trained over three types of tumors: meningioma, glioma and pituitary tumor, on sagittal, coronal and axial views in addition to MRI of normal patients. The proposed model consists of combining a set of several classifiers that uses the features extracted by a convolutional neural network (CNN). We will, also, explore the impact of data augmentation and image resolution on the classification performance with the goal of obtaining the best possible accuracy. We used a CNN architecture and several pre-trained models. The model ResNet 18 gave the highest accuracy of 95.7%.
... Bao et al. [22] proposed a method for detecting abnormal structural health data based on computer vision and deep learning using a fast Fourier transform to convert the time-series data into frequency-domain images; a DNN was designed, trained, and shown to perform well in tests on several types of abnormal data. In terms of improving neural networks for image recognition [23], Gu K [24] designed a wide-depth neural network with both memory and generalization capabilities to learn the nonlinear mapping relationship between features and real PM2.5 concentrations. Gu K [25] designed a deep dual-channel neural network (DCNN) consisting of dual-channel deep sub-networks, which is well applied to smoke detection. ...
The structural health monitoring data of the Research Institute of Highway Ministry of Transportation Track (RIOHTRACK) are huge and complex, including a large amount of dynamic high-frequency sensor data of road underground structures. However, detecting anomalies in the overall distribution of the whole loading cycle data is difficult for traditional numerical data analysis methods. This study proposes an anomaly detection method that visualizes numerical values and designs a deep convolutional neural network DCNN6 for image classification to achieve anomaly detection of large-scale dynamic high-frequency sensor data. After training, the detection rate of DCNN6 for abnormal data reached 92.3% for the validation set. Compared with Residual Neural Network (ResNet50) and GhostNet, the detection accuracy of the method proposed in this study increased by 69% and 4%, respectively, reaching 97%, and the detection speeds were also faster by 5 s/epoch and 4 s/epoch, respectively. Therefore, the proposed method can accurately and quickly detect the abnormality of the dynamic high-frequency sensor data of underground structures, which can provide data support for quickly discovering that the vehicle deviates from the preset trajectory, rectifying the driver's driving deviation, analyzing the force of the whole road area, and grasping the evolution law of the rut.
... Our proposed PAPR reduction approach is implemented in MATLAB on an Intel Pentium 4, 2.33 GHz PC. It is compared to the state-of-the-art OFDM [30], PC-SLM [31], PC-SLM [20], and PC-SLM [7] approaches interms of data rate loss, BLER, BER, SER, and PAPR. Several segments are required to run a MIMO-OFDM system. ...
Wireless communication technologies are playing a very important role in the modern world. The use of multiple antennas with orthogonal frequency division multiple access (MIMO-OFDM) has enhanced spectral efficiency and coverage area. The transmission channel in MIMO-OFDM systems changes with time, creating Inter Symbol Interference (ISI) and Channel Impulse Response (CIR). Modern MIMO-OFDM systems have a high Peak-to-Average Power Ratio (PAPR). Recently, numerous strategies have been developed to minimize PAPR by increasing phase factors. Due to nonlinear effects, high PAPR causes system deterioration. In this paper, we propose hybrid optimal MMSE-MLSE equalizers to enhance PAPR performance in MIMO-OFDM system. First, we introduce hybrid channel equalizer called MMSE and MLSE at the receiver end to improve the performance and to achieve a maximum diversity order. After that, we develop an Adaptive Fruit fly based Deep Neural Network (AF-DNN) by Adaptive Fruit fly Optimization (FOA) to optimally select the equalizer by using the design metrics. Finally, the performance of proposed PAPR reduction using hybrid optimal MMSE-MLSE equalizer is evaluated and their simulation results are compared with the existing state-of-art PAPR reduction techniques in terms of data rate loss, PAPR, BLER, BER and SER.
... A commonplace solution is to prepare and annotate new images, but this is often difficult due to the cost. Thus 1 The University of Tokyo, Tokyo, Japan 2 CyberAgent, Inc., Tokyo, Japan is drawing a great deal of attention [35], and this concept is used for few-shot learning [10] and multi-objective optimization [15]. ...
Transfer learning is a method for improving generalization performance by training a model for a different task first and then additionally training the pre-learned weights for the target task. However, transferability—the ease with which source task can be effectively transferred to which target task—is often unknown. Existing works proposed methods of measuring the transferability between classification tasks using images and discrete labels, but it cannot be applied to regression tasks. In this work, we investigate transferability among classification and regression tasks, and propose a method for predicting transferability by extending the optimal transport theory. Our transferability prediction model also can be applied to subjective tasks (e.g., aesthetics and memorability), which are usually regression tasks. We show that the appropriate source (pre-training) tasks can be predicted for the chosen target task without conducting actual pre-training and transferring trials. Experimental results demonstrated high prediction accuracy (correlation coefficient of ρ=0.791) and a speed improvement of approximately 300 times compared with the above-mentioned greedy approach.
... With their remarkable outstanding feature representation capability, convolutional networks have revolutionized different fields, including the computer vision field [5,47], the industrial field [28], and the monitoring field [27,29]. Recently, segmentation algorithms based on convolutional neural networks (CNNs) have demonstrated SOTA performance for automated biomedical image segmentation [23,45,62]. ...
The process of delineating a region of interest or an object in an image is called image segmentation. Efficient medical image segmentation can contribute to the early diagnosis of illnesses, and accordingly, patient survival possibilities can be enhanced. Recently, deep semantic segmentation methods demonstrate state-of-the-art (SOTA) performance. In this paper, we propose a generic novel deep medical segmentation framework, denoted as Ψnet. This model introduces a novel parallel encoder-decoder structure that draws up the power of triple U-Nets. In addition, a multi-stage squeezed-based encoder is employed to raise the network sensitivity to relevant features and suppress the unnecessary ones. Moreover, atrous spatial pyramid pooling (ASPP) is employed in the bottleneck of the network which helps in gathering more effective features during the training process, hence better performance can be achieved in segmentation tasks. We have evaluated the proposed Ψnet on a variety of challengeable segmentation tasks, including colonoscopy, microscopy, and dermoscopy images. The employed datasets include Data Science Bowl (DSB) 2018 challenge as a cell nuclei segmentation from microscopy images, International Skin Imaging Collaboration (ISIC) 2017 and 2018 as skin lesion segmentation from dermoscopy images, Kvasir-SEG, CVC-ClinicDB, ETIS-LaribDB, and CVC-ColonDB as polyp segmentation from colonoscopy images. Despite the variety in the employed datasets, the proposed model, with extensive experiments, demonstrates superior performance to advanced SOTA models, such as U-Net, ResUNet, Recurrent Residual U-Net, ResUNet++, UNet++, BCDU-Net, MultiResUNet, MCGU-Net, FRCU-Net, Attention Deeplabv3p, DDANet, ColonSegNet, and TMD-Unet.
... The MAML was model-agnostic and general, which was suitable for different types of models and different domains, such as regression, classification, and reinforcement learning in few-shot scenarios. MAML has been successfully applied in many fields: fewshot image classification [26], reinforcement learning [18], speech recognition [16], and smoke density recognition [8]. Reptile was developed based on the well-known MAML architecture, more precisely on FOMAML [21] which was the first-order version of MAML. ...
In practical industrial visual inspection tasks, foreign object data are difficult to collect and accumulate, hence few-shot object detection has gradually become the focus of research. It has been observed that industrial foreign objects are often different from natural data and are always fractal objects. Its form is a rough or fragmented geometric shape, and its features are relatively monotonous and difficult to distinguish. Optimization-based meta-learning is a powerful approach to few-shot learning. It updates model weights through a parameter optimization strategy enabling more efficient learning when faced with new tasks with few samples. Therefore, we proposed a gradient scout strategy, which used the intelligent optimization idea to optimize the meta-training outer-loop parallel gradient optimization method to improve the training effect of few-shot fractal object detection. Meanwhile, we proposed a fractal information amplified learning module, which could improve the detection ability of few-shot fractal objects more quickly under the same training period. They formed FLGS (fractal information amplified learning with gradient scout), which was deployed at zero cost. YOLOv7 was advanced to a new industrial fractal object detection model under FLGS. The experimental results on the IGBT surface foreign object dataset showed that our gradient scout strategy was superior to the other eight few-shot meta-learning algorithms. FLGS significantly accelerated the improvement of fractal object detection ability and maintained a high-level mean average precision.
... Since the proposed algorithm is performed in single image, that is sampling points from an image needs to re-perform the algorithm every time, therefore our method shows very high time complexity. Deep learning has strong approximation ability and extensive applications [21][22][23][24]51], therefore we will develop deep reinforcement learning (e.g., DQN) based methods to sample points in images with an effective and fast way in the future. ...
Adaptive sampling and mesh representation of images play an important role in image compression and vectorization. In this paper, a multi-points stochastic gradient multi-armed bandits algorithm, a generalization of the gradient bandit algorithm, is presented to adaptively sample points in images. By modeling the adaptive image sampling as a multi-arm selection decision-making problem, we first propose an efficient action selection strategy based on a parameterized probability distribution, and then define an adaptive reward function according to the restored image of Delaunay triangulation and a feature map function, and the reward function can overcome the sparse reward issue effectively. As a result, the proposed multi-points stochastic gradient multi-armed bandits algorithm is used to evaluate the reward of each action. At last, a prescribed number of sampling points are selected using a simple and effective strategy according to the average reward of each pixel. The quality of reconstructed images based on sampled points is estimated, and experimental results demonstrate the proposed algorithm achieves a better reconstruction accuracy than that of existing methods.
... Only CO 2 comprises a small portion of the total anthropogenic impact on the environment (Gu et al. 2021). On the other hand, GHG is made up of a variety of emissions such as CO 2 , CH 4 , N 2 O, and SF 6 , and it might be a beneficial indication to exchange CO 2 in appraising EQ (Gu et al. 2018;Gu et al. 2020). Previous research tested negative, positive or neutral signs of parameters throughout the whole sample data. ...
The global outbreak of COVID-19 caused serious threats to public health and economic growth all around the world, but on the other hand, the betterment of the environment took place. How pandemics’ health uncertainty will affect environmental quality is a crucial matter to address. The paper investigates the asymmetric association between pandemics-related health uncertainty and greenhouse gas emissions (GHG) in the top emitter European Union economies (Italy, Germany, France, Poland, Netherlands, Spain, Czech Republic, Belgium, Romania, and Greece). Employing data from 1996 to 2019, a unique approach called ‘Quantile-on-Quantile’, is adopted to evaluate the influence of various quantiles of the health uncertainty on GHG emissions. According to estimates, health uncertainty enhances environmental quality by minimizing GHG in most of our chosen nations at certain quantiles of data, which makes pandemics a blessing in disguise for environmental quality. Additionally, the estimations indicate that the grades of asymmetry between our variables varies by locality, accentuating the requisite for authorities to give specific consideration while executing health uncertainty and environmental quality policies.
... Feature extraction and selection are crucial steps in a pattern or image classification. Their importance is demonstrated in various applications, including PM2.5 feature recognition and air pollution control [27][28][29]. ...
The elastic light-scatter (ELS) technique, which detects and discriminates microbial organisms based on the light-scatter pattern of their colonies, has demonstrated excellent classification accuracy in pathogen screening tasks. The implementation of the multispectral approach has brought further advantages and motivated the design and validation of a hyperspectral elastic light-scatter phenotyping instrument (HESPI). The newly developed instrument consists of a supercontinuum (SC) laser and an acousto-optic tunable filter (AOTF). The use of these two components provided a broad spectrum of excitation light and a rapid selection of the wavelength of interest, which enables the collection of multiple spectral patterns for each colony instead of relying on single band analysis. The performance was validated by classifying microflora of green-leafed vegetables using the hyperspectral ELS patterns of the bacterial colonies. The accuracy ranged from 88.7% to 93.2% when the classification was performed with the scattering pattern created at a wavelength within the 473–709 nm region. When all of the hyperspectral ELS patterns were used, owing to the vastly increased size of the data, feature reduction and selection algorithms were utilized to enhance the robustness and ultimately lessen the complexity of the data collection. A new classification model with the feature reduction process improved the overall classification rate to 95.9%.
... Instead of conducting extensive experiments to heuristically determine the appropriate values of the algorithm's parameters, a deep learning algorithm is capable of automatically learning the high-level features from data in an incremental manner. The adoption of deep learning-based techniques had been gaining in importance because of the impressive performance exhibited in various applications [22][23][24][25][26][27]. The deep learning method in the micro-expression recognition system was first utilized by [28]. ...
Micro-expression is an expression that reveals one’s true feelings and can be potentially applied in various domains such as healthcare, safety interrogation, and business negotiation. The micro-expression recognition is thus far being judged manually by psychologists and trained experts, which consumes a lot of human effort and time. Recently, the development of the deep learning network has proven promising performance in many computer vision-related tasks. Amongst, micro-expression recognition adopts the deep learning methodology to improve the feature learning capability and model generalization. This paper introduces a lightweight apex-based enhanced network that improves by extending one of the state-of-the-art, shallow triple stream three-dimensional CNN. Concretely, the network is first pre-trained with a macro-expression dataset to encounter the small data problem. The features were extracted from CASME II, SMIC, and SAMM datasets. Moreover, thorough recognition results comparison of the datasets are the optical flow-guided features. Besides, an eye masking technique is introduced to reduce noise interference such as eye blinking and glasses reflection issues. The results obtained have an accuracy of 79.19% and an F1-score of 75.9%. Comprehensive experimentation had been conducted on the composite dataset that consists of is provided by comparing it with recent methods. Detailed qualitative and quantitative results are reported and discussed.
... Gu et al. [17] used information richness measures to extract PM2.5-related high-level features. Gu et al. [16] investigated the feasibility of ensemble meta-learning to improve the extraction of key features from deep neural networks through fine-tuning. ...
The emergence of unknown diseases is often with few or no samples available. Zero-shot learning and few-shot learning have promising applications in medical image analysis. In this paper, we propose a Cross-Modal Deep Metric Learning Generalized Zero-Shot Learning (CM-DML-GZSL) model. The proposed network consists of a visual feature extractor, a fixed semantic feature extractor, and a deep regression module. The network belongs to a two-stream network for multiple modalities. In a multi-label setting, each sample contains a small number of positive labels and a large number of negative labels on average. This positive-negative imbalance dominates the optimization procedure and may prevent the establishment of an effective correspondence between visual features and semantic vectors during training, resulting in a low degree of accuracy. A novel weighted focused Euclidean distance metric loss is introduced in this regard. This loss not only can dynamically increase the weight of hard samples and decrease the weight of simple samples, but it can also promote the connection between samples and semantic vectors corresponding to their positive labels, which helps mitigate bias in predicting unseen classes in the generalized zero-shot learning setting. The weighted focused Euclidean distance metric loss function can dynamically adjust sample weights, enabling zero-shot multi-label learning for chest X-ray diagnosis, as experimental results on large publicly available datasets demonstrate.
... Chowdhury et al. [19] proposed a deep CNN to extract feature representations, which were fed to the integration of a support vector machine and one-nearest neighbor classifier for identifying intrusions. To address the flare soot density few shot problem, Gu et al. [20] designed a recognition network for industrial safety and environment protection. To handle the imbalanced data problem, Huang et al. [8] constructed a few-shot learning model with a gated network structure to detect new anomalies by aggregating seen anomaly types and unseen types. ...
The rapid development of Industry 4.0 has amplified the scope and destructiveness of industrial Cyber-Physical System (CPS) by network attacks. Anomaly detection techniques are employed to identify these attacks and guarantee the normal operation of industrial CPS. However, it is still a challenging problem to cope with scenarios with few labeled samples. In this paper, we propose a few-shot anomaly detection model (FSL-PN) based on prototypical network and contrastive learning for identifying anomalies with limited labeled data from industrial CPS. Specifically, we design a contrastive loss to assist the training process of the feature extractor and learn more fine-grained features to improve the discriminative performance. Subsequently, to tackle the overfitting issue during classifying, we construct a robust cost function with a specific regularizer to enhance the generalization capability. Experimental results based on two public imbalanced datasets with few-shot settings show that the FSL-PN model can significantly improve F1 score and reduce false alarm rate (FAR) for identifying anomalous signals to guarantee the security of industrial CPS.
... Optimization algorithms commonly used for this purpose include SGD (Stochastic Gradient Descent), Adam (Adaptive Moment Optimization), RMSProp (Root Mean Square Propagation), SGD + Momentum, Adagrad and Adadelta. First-order iterative algorithms are the most widely used optimizers, originating from gradient descent (Gu et al., 2020a(Gu et al., , 2021a. Let us assume that w is the column vectors that store all the weights of the network. ...
This study aims to develop an integrated approach of deep learning convolutional neural network (DL-CNN) and Google Earth Engine (GEE) platform for salt storm modeling and monitoring. First, we selected several ST's predisposing factors, including Land Surface Temperature (LST), soil salinity, AOD, NDWI and NDVI to train models. We then collected 957 Ground Control Points (GCPs) from the study area, which were randomly divided into training (70%) and validation (30%) datasets. Finally, ReLu, Cross-Entropy, and Adam employed as activation function, loss function and optimizer, respectively. Our findings demonstrate the efficiency of an integrated DL-CNN and GEE for monitoring salt storms (Overall Accuracy (OA) = 0.93.02, 0.92.99, 0.93.88, and 0.92.01 for years 2002, 2010, 2015 and 2021, respectively). The results also show an increase in the frequency of salt storm in the study area from 2002 to 2021. Such approach is a promising step toward understanding, controlling, and managing salt storms and recommend salt storm spatial monitoring in other favored areas with similar environmental conditions. In addition, the results of this study provide critical insights into the environmental impacts of the Lake Urmia drought and its intensive environmental impacts on the human health and wellbeing of the residents. © 2023 Turkish National Committee for Air Pollution Research and Control
Smoke and fire recognition are of great importance on foreseeing fire disasters and preventing environmental pollution by monitoring the burning process of objects (e.g., straw, fuels). However, since fire images suffer from problems like the variability of the features, complexity of scenarios, interference from background, changeable weather conditions as well as image quality problems, identifying smoke and fire accurately and promptly from a given image still remains a substantial challenge. Automatically learning the features of smoke images by CNNs has improved the target recognition ability compared to traditional approaches, nonetheless, convolutions and pooling operations in CNNs may cause severe information loss which may lead to misjudgment. To tackle the above problems, this paper proposed a hybrid attention model based on the characteristics of smoke images. This model adopted multiple optimized attention mechanism in several stages to quickly and precisely capture the important features, achieving state-of-the-art performance on smoke and fire recognition in terms of accuracy and speed. Our proposed module mainly consists of two stages: pooling and attention. In the first stage, we conducted several newly proposed first-order pooling methods. Through traversing the data space in a larger scope, features are better reserved, thus constructing a more intact feature space of smoke and fire in an image. In the second stage, feature maps are aggregated together to perform channel and spatial attention. The channel and spatial dependencies allow us to quickly catch the important features presented in an image. By fully exploring the feature space and prominent salient features, characteristics of smoke and fire are better presented so as to obtain better smoke and fire detection results. Experiments have been conducted on public smoke detection dataset and new proposed fine-grained smoke and fire detection database. Experimental results revealed that the proposed method outperformed popular deep CNNs and existing prevalent attention models for smoke and fire detection problems.
Federated Learning (FL) plays a pivotal role in enabling artificial intelligence (AI)-based mobile applications in mobile edge computing (MEC). However, due to the resource heterogeneity among participating mobile users (MUs), delayed updates from slow MUs may deteriorate the learning speed of the MEC-based FL system, commonly referred to as the straggling problem. To tackle the problem, this work proposes a novel privacy-preserving FL framework that utilizes homomorphic encryption (HE) based solutions to enable MUs, particularly resource-constrained MUs, to securely offload part of their training tasks to the cloud server (CS) and mobile edge nodes (MENs). Our framework first develops an efficient method for packing batches of training data into HE ciphertexts to reduce the complexity of HE-encrypted training at the MENs/CS. On that basis, the mobile service provider (MSP) can incentivize straggling MUs to encrypt part of their local datasets that are uploaded to certain MENs or the CS for caching and remote training. However, caching a large amount of encrypted data at the MENs and CS for FL may not only overburden those nodes but also incur a prohibitive cost of remote training, which ultimately reduces the MSP’s overall profit. To optimize the portion of MUs’ data to be encrypted, cached, and trained at the MENs/CS, we formulate an MSP’s profit maximization problem, considering all MUs’ and MENs’ resource capabilities and data handling costs (including encryption, caching, and training) as well as the MSP’s incentive budget. We then show that the problem is convex and can be efficiently solved using an interior point method. Extensive simulations on a real-world human activity recognition dataset show that our proposed framework can achieve much higher model accuracy (improving up to 24.29%) and faster convergence rate (by 2.86 times) than those of the conventional
FedAvg
approach when the straggling probability varies between 20% and 80%. Moreover, the proposed framework can improve the MSP’s profit up to 2.84 times compared with other baseline FL approaches without MEN-assisted training.
Post-monsoon and winter months cater to conditions such as temperature inversions, crop-residue burning, and firecracker bursting, which enhance air pollution. Utilizing this scenario, this work uses a fuzzy set theory-based approach to explore the interrelationship between different pairs of pollutants. Conducted over Kolkata during the post-monsoon and winter months, the study comprises PM2.5, PM10, NOx, SO2, O3, and temperature as the random variables. Through the computation of descriptive statistics, hypothesis testing through the chi-square test, and quantification of uncertainty by Shannon entropy, we found a lack of symmetry and the presence of uncertainty in the data. In this view, we have considered binary fuzzy relation (BFR) in its raw form and through max–min composition on the random variables to analyze the interrelationship between the pollutants and analyzed the results accordingly. Projections of binary fuzzy relations have supplemented the results. The study concluded that temperature and SO2 can play a significant role in modeling PM2.5, including the effect of PM10.
Ensemble methods are widely used to tackle class imbalance problem. However, for existing imbalanced ensemble (IE) methods, the samples in each subset are resampled from the same dataset, and are directly input to the classifier for training, so the quality (diversity and separability) of the subsets is unsatisfactory usually. To solve the problem, a deep fuzzy envelope sample generation mechanism is proposed. First, the Fuzzy C-Means clustering based deep sample envelope pre-network (DSEN) is designed to mine correlation information among samples, thereby increasing the quality of the subsets. Second, the local manifold structure metric (LMSM) and global structure distribution metric (GSDM) are designed to construct local-global structure consistency mechanism (LGSCM) to enhance distribution consistency of interlayer samples of DSEN. Third, the DSEN and LGSCM are combined to form the final deep sample envelope network–DSENLG to refresh the existing subsets. Finally, base classifiers are applied on the new subsets generated by the DSENLG and then fused, thereby realizing a new IE algorithm. The experimental results show that the proposed algorithm is significantly better than existing representative IE algorithms and it achieves the highest improvement of 10.64%, 19.5%, 18.67% and 22.33% on four criteria over the state-of-the-art methods. The originality of the paper is threefold: (a) proposing the concept of “deep fuzzy samples” or “envelope samples”, which comprehensively considers the correlation information among original samples; (b) proposing the LGSCM to resolve the distribution inconsistency of interlayer samples; and (c) forming an fuzzy envelope sample based IE algorithm.
This research aims at presenting a practical framework for facilitating the identification of development priorities in metropolises. To do so, the CASBEE-city model was applied for a comprehensive quality assessment of the case study, Tehran Metropolis, which covers the environmental, social, and economic aspects and the pollution load. According to the model, Tehran City’s quality is ranked B− or poor. To anticipate the future status of the city, the indicators with the lowest scores were selected and their probable conditions were scored in a Cross Impact Matrix. Considering the existing upstream document for Tehran Development, the most probable scenario for future quality was selected, according to which development measures based on their importance and achievability in the mid-term were prioritized. The results show mitigating the pollution load and improving the social services are among the top priorities in Tehran City, which can favourably affect the city’s status regarding comprehensive quality and sustainable development goals.
Ethylene leakage detection has become one of the most important research directions in the field of target detection due to the fact that ethylene leakage in the petrochemical industry is closely related to production safety and environmental pollution. Under infrared conditions, there are many factors that affect the texture characteristics of ethylene, such as ethylene concentration, background, and so on. We find that the detection criteria used in infrared imaging ethylene leakage detection research cannot fully reflect real-world production conditions, which is not conducive to evaluate the performance of current image-based target detection methods. Therefore, we create a new infrared image dataset of ethylene leakage with different concentrations and backgrounds, including 54275 images. We use the proposed dataset benchmark to evaluate seven advanced image-based target detection algorithms. Experimental results demonstrate the performance and limitations of existing algorithms, and the dataset benchmark has good versatility and effectiveness.
Smoke detection plays a crucial role in the safety production of petrochemical enterprises and fire prevention. Image-based machine learning and deep learning methods have been widely studied. Recently, many works have applied the transformer to solve problems faced by computer vision tasks (such as classification and object detection). To our knowledge, there are few studies using the transformer structure to detect smoke. In order to research the application potential and improve the performance of the transformer in the smoke detection field, we propose a model consisting of two transformer encoders and a convolutional neural network (CNN) module. The first transformer encoder can be used to establish the global relationship of an image, and the CNN structure can provide additional local information to the transformer. The fusion of global information and local information is conducive to the second transfer encoder to make better decisions. Experiments results on large-size dataset for industrial smoke detection illustrate the effectiveness of the proposed model.
In the era of media information explosion, there is an urgent need for a fast and reliable image quality assessment (IQA) model to improve the actual application effect of images. To this end, we propose the multiple information measurement fusion metric (MMFM), which innovatively combines two types of information measures (IMs), i.e., local IM and global IM, using only a small number of references for IQA. First, inspired by the free energy theory, we combine 2-dimensional autoregressive model with sparse random sampling method as an inference engine on an input image to generate its associated predicted image. Second, by the inspiration of pixel-wise measurement, we obtain the local IM by calculating the information entropy of the residual error between the input image and its corresponding predicted one. Third, motivated by the histogram-based measurement, we acquire the global IM by computing the two kinds of divergences between the input image and its predicted one. Fourth, we systematically fuse three components, independently including one distance of local IM between the reference and corrupted images and two distances of global IMs between the reference and corrupted images, based on a linear function to derive the final IQA result. The results of experiment on the most popular LIVE database show that our designed algorithm with only one number used as few reference has achieved well performance as compared with several mainstream IQA models.
River turbidity is an important index to evaluate the quality of water environment, which is of great significance to environmental monitoring. Most of the existing monitoring methods rely on turbidity labels measured by river water quality monitoring stations. However, due to the limitation of spatial distribution of water quality monitoring stations, those turbidity data outside the monitoring range are difficult to obtain and the correlation of turbidity data between monitoring stations is poor. Using such data directly might weaken the generalization ability that the model has. In this paper, we propose a new river turbidity monitoring model based on semi-supervised transfer learning (RTMM-SSTL) to solve these problems. First, in the pre-training stage, we innovatively propose a semi-supervised training method combined with big data with pseudo labels for river turbidity monitoring to solve the problem of weakened model generalization ability. Then, the model is fine-tuned using GEE hyper spectral data with ground truth to further improve the monitoring capability of the model. Experiments conducted on the river turbidity monitoring task demonstrate that the proposed model is superior to the advanced learning models, and further proves that our semi-supervised transfer learning method performs better than the state-of-the-art supervision models.
Few-shot data collected from hoisting system suffers from inadequate information in the practical industries, which reduces the diagnostic accuracy of the data-driven based fault diagnosis approaches. To overcome this problem, a self-supervised meta learning generative adversarial network (SSMLGAN) algorithm is proposed in this paper. The purpose of the proposed algorithm is to determine the optimal initialization parameters of the model by training on various data generation tasks, thus accomplishing new data generation using only a small amount of training data. Specifically, a self-supervised strategy is proposed to improve the generalization performance of the proposed algorithm. Besides, the fault data of hoisting system is collected for data generation, and the experimental results show that the proposed algorithm can determine the optimal initialization parameters under the condition of insufficient datasets. The effectiveness of the proposed algorithm for few-shot fault diagnosis is verified by using a mixture of real data and generated data.
Air pollutants and allergens are the main stimuli that have considerable effects on asthmatic patients’ health. Seamless monitoring of patients’ conditions and the surrounding environment, limiting their exposure to allergens and irritants, and reducing the exacerbation of symptoms can aid patients to deal with asthma better. In this context, ubiquitous healthcare monitoring systems can provide any service to any user everywhere and every time through any device and network. In this regard, this research established a GIS-based outdoor asthma monitoring framework in light of ubiquitous systems. The proposed multifaceted model was designed in three layers: (1) pre-processing, for cleaning and interpolating data, (2) reasoning, for deducing knowledge and extract contextual information from data, and (3) prediction, for estimating the asthmatic conditions of patients ubiquitously. The effectiveness of the proposed model is assessed by applying it on a real dataset that comprised of internal context information including patients’ personal information (age, gender, height, medical history), patients’ locations, and their peak expiratory flow (PEF) values, as well as external context information including air pollutant data (O3, SO2, NO2, CO, PM10), meteorological data (temperature, pressure, humidity), and geographic information related to the city of Tehran, Iran. With more than 92% and 93% accuracies in reasoning and estimation mechanism, respectively, the proposed method showed remarkably effective in asthma monitoring and management.
Recent work has shown that convolutional networks can be substantially deeper, more accurate and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper we embrace this observation and introduce the Dense Convolutional Network (DenseNet), where each layer is directly connected to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections, one between each layer and its subsequent layer (treating the input as layer 0), our network has L(L+1)/2 direct connections. For each layer, the feature maps of all preceding layers are treated as separate inputs whereas its own feature maps are passed on as inputs to all subsequent layers. Our proposed connectivity pattern has several compelling advantages: it alleviates the vanishing gradient problem and strengthens feature propagation; despite the increase in connections, it encourages feature reuse and leads to a substantial reduction of parameters; its models tend to generalize surprisingly well. We evaluate our proposed architecture on five highly competitive object recognition benchmark tasks. The DenseNet obtains significant improvements over the state-of-the-art on all five of them (e.g., yielding 3.74% test error on CIFAR-10, 19.25% on CIFAR-100 and 1.59% on SVHN).
It is a challenging task to recognize smoke from images due to large variance of smoke color, texture and shapes. There are smoke detection methods that have been proposed, but most of them are based on hand-crafted features. To improve the performance of smoke detection, we propose a novel deep normalization and convolutional neural network (DNCNN) with fourteen layers to implement automatic feature extraction and classification. In DNCNN, traditional convolutional layers are replaced with normalization and convolutional layers to accelerate the training process and boost the performance of smoke detection. To reduce overfitting caused by imbalanced and insufficient training samples, we generate more training samples from original training data sets by using a variety of data enhancement techniques. Experimental results show that our method achieved very low false alarm rates below 0.60% with detection rates above 96.37% on our smoke datasets.
Recent work has shown that convolutional networks can be substantially deeper, more accurate, and efficient to train if they contain shorter connections between layers close to the input and those close to the output. In this paper, we embrace this observation and introduce the Dense Convolutional Network (DenseNet), which connects each layer to every other layer in a feed-forward fashion. Whereas traditional convolutional networks with L layers have L connections - one between each layer and its subsequent layer - our network has L(L+1)/2 direct connections. For each layer, the feature-maps of all preceding layers are used as inputs, and its own feature-maps are used as inputs into all subsequent layers. DenseNets have several compelling advantages: they alleviate the vanishing-gradient problem, strengthen feature propagation, encourage feature reuse, and substantially reduce the number of parameters. We evaluate our proposed architecture on four highly competitive object recognition benchmark tasks (CIFAR-10, CIFAR-100, SVHN, and ImageNet). DenseNets obtain significant improvements over the state-of-the-art on most of them, whilst requiring less memory and computation to achieve high performance. Code and models are available at https://github.com/liuzhuang13/DenseNet.
Flaring is a high-temperature oxidation process used to burn waste gases from industrial operations. Smoke results from combustion, depends on waste gas components, quantity and distribution of combustion air. Flares stacks are used in industries are often assisted with steam to ensure complete combustion and to avoid any unburnt hydro carbons. In this process, flares are often sooty due to insufficient steam. This results in black carbon from flares. Soot particles in the air are a contributing factor in respiratory diseases. The fine particles less than 3micron are the worst causes of lung damage due to their ability to penetrate into the deep air passage. This paper explains the health effects of soot and particulate matter. New scientific evidence has led to recognition of the significant role of black particles (black carbon – BC) as one of the short-lived climate forcers. Measures focused on BC and methane is expected to achieve a significant short-term reduction in global warming. The black carbon or soot from flares can be minimized by controlling steam to flare by automation. Flare gas flow measurement by ultrasonic flow meter gives a high turn down ratio of 2000:1. This paper explains the methodology of control of soot from flares by steam to hydro carbon ratio control and how the combustion efficiency varies with the amount of steam. The economical benefit of saving steam by automation is not only credit to the company but also carbon credit to the world.
This paper addresses the problem of matching common node correspondences among multiple graphs referring to an identical or related structure. This multi-graph matching problem involves two correlated components: i) the local pairwise matching affinity across pairs of graphs; ii) the global matching consistency that measures the uniqueness of the pairwise matchings by different composition orders. Previous studies typically either enforce the matching consistency constraints in the beginning of an iterative optimization, which may propagate matching error both over iterations and across graph pairs; or separate affinity optimization and consistency enforcement into two steps. This paper is motivated by the observation that matching consistency can serve as a regularizer in the affinity objective function especially when the function is biased due to noises or inappropriate modeling. We propose composition-based multi-graph matching methods to incorporate the two aspects by optimizing the affinity score, meanwhile gradually infusing the consistency. We also propose two mechanisms to elicit the common inliers against outliers. Compelling results on synthetic and real images show the competency of our algorithms.
Several machine learning models, including neural networks, consistently
misclassify adversarial examples---inputs formed by applying small but
intentionally worst-case perturbations to examples from the dataset, such that
the perturbed input results in the model outputting an incorrect answer with
high confidence. Early attempts at explaining this phenomenon focused on
nonlinearity and overfitting. We argue instead that the primary cause of neural
networks' vulnerability to adversarial perturbation is their linear nature.
This explanation is supported by new quantitative results while giving the
first explanation of the most intriguing fact about them: their generalization
across architectures and training sets. Moreover, this view yields a simple and
fast method of generating adversarial examples. Using this approach to provide
examples for adversarial training, we reduce the test set error of a maxout
network on the MNIST dataset.
This work deals with the multi-faceted impact of gas flaring on a global scale and the different approach employed by researchers to measure gas flared and its resulting emissions. It gives an overview of methods employed by these re-searchers in the oil and gas industry, academia and governments in attempt to determine ways of measuring and reduc-ing gas flaring and its emission drastically. This approach so far includes analytical studies, numerical studies, modeling, computer simulations, etc. the goal behind each study being to mitigate the effects of gas flaring. The outcome indicates that there is a seemingly absence of a single global method, emission factor and estimation procedure used in the oil and gas industry all over the world to determine the volume of gas flared and its emissions be it from complete or incom-plete combustion, sweet or sulphur present hydrocarbons and this poses a continuous problem in determining the actual impact of gas flaring and its emissions on human and its role in environmental degradation both at a local and global level. An attempt has also being made to cover up-to-date trends in gas flaring and current developments in some of the most flared countries.
We propose a deep convolutional neural network architecture codenamed
"Inception", which was responsible for setting the new state of the art for
classification and detection in the ImageNet Large-Scale Visual Recognition
Challenge 2014 (ILSVRC 2014). The main hallmark of this architecture is the
improved utilization of the computing resources inside the network. This was
achieved by a carefully crafted design that allows for increasing the depth and
width of the network while keeping the computational budget constant. To
optimize quality, the architectural decisions were based on the Hebbian
principle and the intuition of multi-scale processing. One particular
incarnation used in our submission for ILSVRC 2014 is called GoogLeNet, a 22
layers deep network, the quality of which is assessed in the context of
classification and detection.
The flare stack is a typical flare gas combustion facility used to guarantee the safe production of petrochemical plants, refineries and other enterprises. One of the most vital problems of a flare stack is the incomplete combustion of flare gas, which produces a large amount of flare soot and thus endangers the atmosphere and human health. Hence, an effective and efficient flare soot monitoring system that has important guiding significance to environmental protection is strongly required. To this end, we devise a vision-based monitor of flare soot (VMFS) that can search for flare soot in a timely way and ensure the full combustion of flare gas. First, the proposed VMFS leverages the broadly tuned colour channel to recognize a flame in an input video frame since the flame is the source of flare soot in our application. Second, our monitor incorporates fast saliency detection with K-means to fix the position of the flame. Third, we take the flame area as the centre to search for the potential flare soot region, followed by identifying the flare soot based on the background colour channel. The results of experiments on multiple video sequences collected at a real petrochemical plant reveal that the proposed VMFS is superior to state-of-the-art relevant models in both monitoring performance and computational efficiency. The implementation code will soon be released at https://kegu.netlify.com/.
Internet of Things (IoT) has been widely utilized with Artificial Intelligence (AI). However, it requires substantial effort to integrate AI and big data with IoT. To mitigate this problem, AItalk was proposed. By treating AI as a cyber IoT device, we do not need to write code of the AI mechanism in the network applications, as traditional AI-based IoT applications did. This paper describes how the AI tools such as scikit-learn and TensorLayer are accommodated as cyber IoT devices in AItalk, and extends AItalk for non-IoT applications. We use house valuation as an example to show how AItalk can flexibly include the factors that have significant impact on the house price. We show that by adding extra features other than housing profile features, the accuracy for the prediction (valuation) can be improved by 40%. We also investigated the communication overhead of the distributed AItalk structure, which is 3.7% for the computation of one house price valuation.
Artificial-intelligence researchers are trying to fix the flaws of neural networks. Artificial-intelligence researchers are trying to fix the flaws of neural networks.
PM
2.5
is an important global air pollution issue. PM
2.5
causes various problems on the environment and human health. Even though PM
2.5
of the outdoor air is eliminated by particulate air filters, the chemicals adsorbed on the filtered PM
2.5
can be outgassed and introduced into the semiconductor manufacturing facilities and processes, thus causing defects on the semiconductor chips by producing particles and haze as airborne molecular contamination. Therefore, estimating the amount of the outgassing from PM
2.5
is a very important factor to control gaseous contaminants in the semiconductor industry. In this paper, the PM
2.5
samples at different PM
2.5
mass loadings were collected in Xi'an, China and evaluated using the soft X-ray radiolysis method for determining the relation between the outgassing and the PM
2.5
mass loading. The PM
2.5
samples were also analyzed by several methods for quantifying organic and elemental carbons, water soluble ions, and organic compounds in the PM
2.5
samples. The outgassing measured by the soft X-ray method increased linearly with the PM
2.5
mass loading, and the chemical analyses explained that the organic carbon (OC) was dominantly detected by the soft X-ray method in the current study. Moreover, the sampling location of the PM
2.5
affected on the organic compounds of the PM
2.5
.
Accurate measurement of position and attitude information is particularly important. Traditional measurement methods generally require high-precision measurement equipment for analysis, leading to high costs and limited applicability. Vision-based measurement schemes need to solve complex visual relationships. With the extensive development of neural networks in related fields, it has become possible to apply them to the object position and attitude. In this paper, we propose an object pose measurement scheme based on convolutional neural network and we have successfully implemented end-to-end position and attitude detection. Furthermore, to effectively expand the measurement range and reduce the number of training samples, we demonstrated the independence of objects in each dimension and proposed subadded training programs. At the same time, we generated generating image encoder to guarantee the detection performance of the training model in practical applications.
Smoke detection plays an important role in industrial safety warning systems and fire prevention. Due to the complicated changes in the shape, texture and colour of smoke, identifying the smoke from a given image still remains a substantial challenge, and this has accordingly aroused a considerable amount of research attention recently. To address the problem, we devise a new deep dual-channel neural network (DCNN) for smoke detection. In contrast to popular deep convolutional networks, e.g., Alex-Net, VGG-Net, Res-Net, and Dense-Net, and the DNCNN that is specifically devoted to detecting smoke, our proposed end-to-end network is mainly composed of dual channels of deep subnetworks. In the first subnetwork, we sequentially connect multiple convolutional layers and max-pooling layers. Then, we selectively append the batch normalization layer to each convolutional layer for over-fitting reduction and training acceleration. The first subnetwork is shown to be good at extracting the detail information of smoke, such as texture. In the second subnetwork, in addition to the convolutional, batch normalization and max-pooling layers, we further introduce two important components. One is the skip connection for avoiding the vanishing gradient and improving the feature propagation. The other is the global average pooling for reducing the number of parameters and mitigating the over-fitting issue. The second subnetwork can capture the base information of smoke, such as contours. We finally deploy a concatenation operation to combine the aforementioned two deep subnetworks to complement each other. Based on the augmented data obtained by rotating the training images, our proposed DCNN can promptly and stably converge to the perfect performance. Experimental results conducted on the publicly available smoke detection database verify that the proposed DCNN has attained a very high detection rate that exceeds 99.5% on average, superior to state-of-the-art relevant competitors. Furthermore, our DCNN only employs approximately one-third of the parameters needed by the comparatively tested deep neural networks. The source code of DCNN will be released to the public.
Most of the current blind stereoscopic image quality assessment (SIQA) algorithms cannot show reliable accuracy. One reason is that they do not have the deep architectures and the other reason is that they are designed on the relatively weak biological basis, compared with findings on human visual system (HVS). In this paper, we propose a Deep Edge and COlor Signal INtegrity Evaluator (DECOSINE) based on the whole visual perception route from eyes to the frontal lobe, and especially focus on edge and color signal processing in retinal ganglion cells (RGC) and lateral geniculate nucleus (LGN). Furthermore, to model the complex and deep structure of the visual cortex, Segmented Stacked Auto-encoder (S-SAE) is used, which has not utilized for SIQA before. The utilization of the S-SAE complements weakness of deep learning-based SIQA metrics that require a very long training time. Experiments are conducted on popular SIQA databases, and the superiority of DECOSINE in terms of prediction accuracy and monotonicity is proved. The experimental results show that our model about the whole visual perception route and utilization of S-SAE are effective for SIQA.
Land use regression (LUR) is commonly used to estimate air pollution exposures for epidemiological studies. By statistically relating a set of geolocated measured pollutant values with explanatory variables defining sources and modifiers of air pollution patterns, such as land cover characteristics, traffic flow and intensity, it is possible to predict pollution levels at unsampled locations. LUR utilises simple linear regression, but the generation of predictor variables, application of the model and the supervised iterative approach to model development means an analyst must be a competent user of both GIS and statistical packages. Here we present an application to simplify the LUR modelling process for exposure scientists and environmental epidemiologists. RLUR is a user-friendly application built using the statistical and GIS capabilities of the R programming language. The main aim of this software is to provide an introduction to the LUR process without the need for specific GIS or statistical expertise.
Blind or visually impaired people face special difficulties in daily life. With the advances in vision sensors and computer vision, the design of wearable vision assistance system is promising. In order to improve the life quality of the visually impaired group, a wearable system is proposed in this paper. Typically the performance of visual sensors is affected by a variety of complex factors in practice, resulting in a large number of noise and distortion. In this paper, we will creatively leverage image quality evaluation to select the captured images through vision sensors, which can ensure the input quality of scenes for the final identification system. First, we use binocular vision sensors to capture images in a fixed frequency and choose the informative ones based on stereo image quality assessment (SIQA). Then the captured images will be sent to cloud for further computing. Specially, the detection and automatic result will be done for all the received images. Convolutional neural network based on big data will be used in this step. According to image analysis, the cloud computing can return the requested information for users, which can help them make a more reasonable decision in further action. Simulations and experiments show that the proposed method can solve the problem effectively. In addition, statistical results also demonstrate that wearable vision system can make visually impaired group more satisfied in visual needed situations.
Ensemble methods are considered the state‐of‐the art solution for many machine learning challenges. Such methods improve the predictive performance of a single model by training multiple models and combining their predictions. This paper introduce the concept of ensemble learning, reviews traditional, novel and state‐of‐the‐art ensemble methods and discusses current challenges and trends in the field.
This article is categorized under:
• Algorithmic Development > Model Combining
• Technologies > Machine Learning
• Technologies > Classification
Few-shot learning remains challenging for meta-learning that learns a learning algorithm (meta-learner) from many related tasks. In this work, we argue that this is due to the lack of a good representation for meta-learning, and propose deep meta-learning to integrate the representation power of deep learning into meta-learning. The framework is composed of three modules, a concept generator, a meta-learner, and a concept discriminator, which are learned jointly. The concept generator, e.g. a deep residual net, extracts a representation for each instance that captures its high-level concept, on which the meta-learner performs few-shot learning, and the concept discriminator recognizes the concepts. By learning to learn in the concept space rather than in the complicated instance space, deep meta-learning can substantially improve vanilla meta-learning, which is demonstrated on various few-shot image recognition problems. For example, on 5-way-1-shot image recognition on CIFAR-100 and CUB-200, it improves Matching Nets from 50.53% and 56.53% to 58.18% and 63.47%, improves MAML from 49.28% and 50.45% to 56.65% and 64.63%, and improves Meta-SGD from 53.83% and 53.34% to 61.62% and 66.95%, respectively.
Aircraft detection in remote sensing images has become an attractive research topic, which plays an essential role in various military and civil applications. In this letter, we develop a novel method for aircraft detection in remote sensing images based on deep residual network (ResNet) and Super-Vector (SV) coding. First, a variant of ResNet with fewer layers is designed to increase the resolution of the feature map, and multi-level convolutional features are merged into an informative feature description for region proposal. Meanwhile, we extract histogram of oriented gradient (HOG) with SV coding from each region of interest, which assists convolutional features to complete object classification. We comprehensively evaluate the proposed method on our remote sensing image dataset. The experimental results show that our method outperforms top-performing aircraft detection methods with higher accuracy even when the backgrounds are complicated.
Several machine learning models, including neural networks, consistently mis- classify adversarial examples—inputs formed by applying small but intentionally worst-case perturbations to examples from the dataset, such that the perturbed in- put results in the model outputting an incorrect answer with high confidence. Early attempts at explaining this phenomenon focused on nonlinearity and overfitting. We argue instead that the primary cause of neural networks' vulnerability to ad- versarial perturbation is their linear nature. This explanation is supported by new quantitative results while giving the first explanation of the most intriguing fact about them: their generalization across architectures and training sets. Moreover, this view yields a simple and fast method of generating adversarial examples. Us- ing this approach to provide examples for adversarial training, we reduce the test set error of a maxout network on the MNIST dataset.
We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 dif- ferent classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make training faster, we used non-saturating neurons and a very efficient GPU implemen- tation of the convolution operation. To reduce overfitting in the fully-connected layers we employed a recently-developed regularization method called dropout that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry
Few-shot learning is challenging for learning algorithms that learn each task in isolation and from scratch. In contrast, meta-learning learns from many related tasks a meta-learner that can learn a new task more accurate and faster with fewer examples, where the choice of meta-learners is crucial. In this paper, we develop an SGD-like, easily trainable meta-learner, called Meta-SGD, that can initialize and adapt any differentiable learner in just one step. Compared to the popular meta-learner LSTM, Meta-SGD is conceptually simple, easy to implement, and can be learned efficiently. Compared to the latest meta-learner MAML, Meta-SGD has a much higher capacity in learning to learn not just the learner initialization, but also the learner update direction and learning rate, all in a single meta-learning process. Meta-SGD shows highly competitive performance for few-shot learning on regression and classification.
We propose an algorithm for meta-learning that is model-agnostic, in the sense that it is compatible with any model trained with gradient descent and applicable to a variety of different learning problems, including classification, regression, and reinforcement learning. The goal of meta-learning is to train a model on a variety of learning tasks, such that it can solve new learning tasks using only a small number of training samples. In our approach, the parameters of the model are explicitly trained such that a small number of gradient steps with a small amount of training data from a new task will produce good generalization performance on that task. In effect, our method trains the model to be easy to fine-tune. We demonstrate that this approach leads to state-of-the-art performance on a few-shot image classification benchmark, produces good results on few-shot regression, and accelerates fine-tuning for policy gradient reinforcement learning with neural network policies.
This paper addresses the problem of hypergraph matching using higher-order affinity information. We propose a solver that iteratively updates the solution in the discrete domain by linear assignment approximation. The proposed method is guaranteed to converge to a stationary discrete solution and avoids the annealing procedure and ad-hoc post binarization step that are required in several previous methods. Specifically, we start with a simple iterative discrete gradient assignment solver. This solver can be trapped in an m-circle sequence under moderate conditions, where m is the order of the graph matching problem. We then devise an adaptive relaxation mechanism to jump out this degenerating case and show that the resulting new path will converge to a fixed solution in the discrete domain. The proposed method is tested on both synthetic and real-world benchmarks. The experimental results corroborate the efficacy of our method.
Recently developed data acquisition equipment and data processing methods have ignited the possibility of power system online sensitivity identification (OSI). Despite the existing OSI algorithms, practical issues such as data collinearity and the noise effect on the identification algorithm must be considered to realize OSI in real power systems. In this study, the negative and positive aspects of noise to OSI are first studied. Then, under the data collinearity condition and by making use of the positive as-pects of noise, a noise-assisted ensemble regression method (NAER) is proposed to simultaneously solve the data collinearity problem and manage the negative aspects of noise. Moreover, the proposed method is proven equivalent to one of the most effective measures, the norm-2 regularization method, to address the col-linearity problem, and therefore provides satisfactory OSI results. The proposed method is tested in an 8-generator 36-node system with original operations data from a real power system, and the results validate its effectiveness.
Using empirical analysis, conventional air automatic monitoring system has high precision, but large bulk, high cost, and single datum class make it impossible for large-scale installation. Based on intriducing Internet of Things(IOT) into the field of environmental protection, this paper puts forward a kind of real-time air pollution monitoring and forecasting system. By using IOT, this system can reduce the hardware cost into 1/10 as before. The system can be laid out in a large number in monitoring area to form monitoring sensor network. Besides the functions of conventional air automatic monitoring system, it also exhibits the function of forecasting development trend of air pollution within a certain time range by analyzing the data obtained by front-end perception system according to neural network technology. Targeted emergency disposal measures can be taken to minimize losses in practical application.
Deeper neural networks are more difficult to train. We present a residual
learning framework to ease the training of networks that are substantially
deeper than those used previously. We explicitly reformulate the layers as
learning residual functions with reference to the layer inputs, instead of
learning unreferenced functions. We provide comprehensive empirical evidence
showing that these residual networks are easier to optimize, and can gain
accuracy from considerably increased depth. On the ImageNet dataset we evaluate
residual nets with a depth of up to 152 layers---8x deeper than VGG nets but
still having lower complexity. An ensemble of these residual nets achieves
3.57% error on the ImageNet test set. This result won the 1st place on the
ILSVRC 2015 classification task. We also present analysis on CIFAR-10 with 100
and 1000 layers.
The depth of representations is of central importance for many visual
recognition tasks. Solely due to our extremely deep representations, we obtain
a 28% relative improvement on the COCO object detection dataset. Deep residual
nets are foundations of our submissions to ILSVRC & COCO 2015 competitions,
where we also won the 1st places on the tasks of ImageNet detection, ImageNet
localization, COCO detection, and COCO segmentation.
The problem of graph matching in general is NPcomplete and many approximate pairwise matching techniques have been proposed. For a general setting in real applications, it typically requires to find the consistent matchings across a batch of graphs. Sequentially performing pairwise matching is prone to error propagation along the pairwise matching sequence, and the sequences generated in different pairwise matching orders can lead to contradictory solutions. Motivated by devising a robust and consistent multiple-graph matching model, we propose a unified alternating optimization framework for multigraph matching. In addition, we define and use two metrics related to graph-wise and pairwise consistencies. The former is used to find an appropriate reference graph which induces a set of basis variables and launches the iteration procedure. The latter defines the order in which the considered graphs in the iterations are manipulated. We show two embodiments under the proposed framework that can cope with the nonfactorized and factorized affinity matrix, respectively. Our multigraph matching model has two major characters: i) the affinity information across multiple graphs are explored in each iteration by fixing part of the matching variables via a consistencydriven mechanism; ii) the framework is flexible to incorporate various existing pairwise graph matching solvers in an "outof- box" fashion, and also can proceed with the output of other multi-graph matching methods. The experimental results on both synthetic data and real images empirically show that the proposed framework performs competitively with the state-of-the-art.
We trained a large, deep convolutional neural network to classify the 1.2 million high-resolution images in the ImageNet LSVRC-2010 contest into the 1000 dif-ferent classes. On the test data, we achieved top-1 and top-5 error rates of 37.5% and 17.0% which is considerably better than the previous state-of-the-art. The neural network, which has 60 million parameters and 650,000 neurons, consists of five convolutional layers, some of which are followed by max-pooling layers, and three fully-connected layers with a final 1000-way softmax. To make train-ing faster, we used non-saturating neurons and a very efficient GPU implemen-tation of the convolution operation. To reduce overfitting in the fully-connected layers we employed a recently-developed regularization method called "dropout" that proved to be very effective. We also entered a variant of this model in the ILSVRC-2012 competition and achieved a winning top-5 test error rate of 15.3%, compared to 26.2% achieved by the second-best entry.
In this work we investigate the effect of the convolutional network depth on
its accuracy in the large-scale image recognition setting. Our main
contribution is a thorough evaluation of networks of increasing depth, which
shows that a significant improvement on the prior-art configurations can be
achieved by pushing the depth to 16-19 weight layers. These findings were the
basis of our ImageNet Challenge 2014 submission, where our team secured the
first and the second places in the localisation and classification tracks
respectively.
Today’s world is facing global warming as one of its main issues. A rise in carbon dioxide and other greenhouse gases concentration in the atmosphere causes this problem. A suggested method for controlling the level of greenhouse gases in the atmosphere is prevention of flaring gas. In this work, three methods are proposed to recover gas instead of conventional burning in a flare at the Asalooye Gas Refinery. These methods aim to minimize environmental and economical disadvantages of burning flare gas. The proposed methods are: 1) Gas-to-Liquid (GTL) production, 2) electricity generation with a gas turbine and, 3) compression and injection into the refinery pipelines. In order to find the most suitable method the required equipments for the three aforementioned methods are simulated. These simulations determine the amount of flare gas, the number of GTL barrels, the power generated by the gas turbine and the required compression horsepower. The results of the simulation show that 48,056 barrels per day of valuable GTL products is produced by the first method. The second method provides 2130 MW electricity and the third method provides a compressed natural gas with 129 bar pressure for injection to the refinery pipelines. In addition, the economics of the flare gas recovery methods are studied and compared. The results show that for the 356.5 MMSCFD of gas flared from the Asalooye Gas Refinery, GTL production provides the greatest rate of return. However, GTL requires the greatest capital investment. Gas compression yields the next highest rate of return and due to the lower capital requirements is the best choice for the Asalooye Gas Refinery. This new process is a suitable alternative to conventional gas flaring which prevents harmful environmental effects through emission of significant amounts of carbon dioxide in the atmosphere.
Rate constants have been measured at 296 ± 2 K for the gas-phase reactions of camphor with OH radicals, NO3 radicals, and O3. Using relative rate methods, the rate constants for the OH radical and NO3 radical reactions were (4.6 ± 1.2) × 10−12 cm3 molecule−1 s−1 and <3 × 10−16 cm3 molecule−1 s−1, respectively, where the indicated error in the OH radical reaction rate constant includes the estimated overall uncertainty in the rate constant for the reference compound. An upper limit to the rate constant for the O3 reaction of <7 × 10−20 cm3 molecule−1 s−1 was also determined. The dominant tropospheric loss process for camphor is calculated to be by reaction with the OH radical. Acetone was identified and quantified as a product of the OH radical reaction by gas chromatography, with a formation yield of 0.29 ± 0.04. In situ atmospheric pressure ionization tandem mass spectrometry (API-MS) analyses indicated the formation of additional products of molecular weight 166 (dicarbonyl), 182 (hydroxydicarbonyl), 186, 187, 213 (carbonyl-nitrate), 229 (hydroxycarbonyl-nitrate), and 243. A reaction mechanism leading to the formation of acetone is presented, as are pathways for the formation of several of the additional products observed by API-MS. © 2000 John Wiley and Sons, Inc. Int J Chem Kinet 33: 56–63, 2001
Emissions of incompletely burned hydrocarbons from industrial flares may contribute to air pollution. Available data on flare emissions are sparse, and methods to sample operating flares are unavailable. The study reported herein was developed to provide additional data on flare emissions. A Flare Test Facility (FTF) was designed and constructed. Tests were conducted on 3-, 6-, and 12-in.† diameter flare heads. Propane was used as the flare fuel, diluted with nitrogen to control the heating value. The following results were obtained: (1) soot (from smoky flares) accounts for less than 0.5 percent of the unburned hydrocarbon emissions; (2) the size of the flare head did not influence hydrocarbon combustion efficiency; (3) the stability of the flare flame influenced combustion efficiency, with unstable flames tending to promote inefficient combustion. A relationship between gas heating value and exit velocity was developed to denote the region of flame instability.
Atmospheric chemistry of VOCs and NOx
- roger
Measurements of outgassing from PM
- C Kim
- C Chen
- J Zhou
- J Cao
- D Y H Pui
A land use regression variable generation, modelling and prediction tool for air pollution exposure assessment
- K C Neda
- A A Ali
- S Mohammad
MxML: Mixture of meta-learners for few-shot classification
- M Park
- J Kim
- S Kim
- Y Liu
- S Choi
MxML: Mixture of meta-learners for few-shot classification
- park