PeerJ Computer Science

Parking space prediction is a significant aspect of smart cities. It is essential for addressing traffic congestion challenges and low parking availability in urban areas. The present research mainly focuses on proposing a novel scalable hybrid model for accurately predicting parking space. The proposed model works in two phases: in first phase, auto-regressive integrated moving average (ARIMA) and long short-term memory (LSTM) models are integrated. Further, in second phase, backpropagation neural network (BPNN) is used to improve the accuracy of parking space prediction by reducing number of errors. The model utilizes the ARIMA model for handling linear values and the LSTM model for targeting non-linear values of the dataset. The Melbourne Internet of Things (IoT) based dataset, is used for implementing the proposed hybrid model. It consists of the data collected from the sensors that are employed in smart parking areas of the city. Before analysis, data was pre-processed to remove noise from the dataset and real time information collected from different sensors to predict the results accurately. The proposed hybrid model achieves the minimum mean squared error (MSE), mean absolute error (MAE), and root mean squared error (RMSE) values of 0.32, 0.48, and 0.56, respectively. Further, to verify the generalizability of the proposed hybrid model, it is also implemented on the Harvard IoT-based dataset. It achieves the minimum MSE, MAE, and RMSE values of 0.31, 0.47, and 0.56, respectively. Therefore, the proposed hybrid model outperforms both datasets by achieving minimum error, even when compared with the performance of other existing models. The proposed hybrid model can potentially improve parking space prediction, contributing to sustainable and economical smart cities and enhancing the quality of life for citizens.

Artificial intelligence (AI) in music improvisation offers promising new avenues for developing human creativity. The difficulty of writing dynamic, flexible musical compositions in real time is discussed in this article. We explore using reinforcement learning (RL) techniques to create more interactive and responsive music creation systems. Here, the musical structures train an RL agent to navigate the complex space of musical possibilities to provide improvisations. The melodic framework in the input musical data is initially identified using bi-directional gated recurrent units. The lyrical concepts such as notes, chords, and rhythms from the recognised framework are transformed into a format suitable for RL input. The deep gradient-based reinforcement learning technique used in this research formulates a reward system that directs the agent to compose aesthetically intriguing and harmonically cohesive musical improvisations. The improvised music is further rendered in the MIDI format. The Bach Chorales dataset with six different attributes relevant to musical compositions is employed in implementing the present research. The model was set up in a containerised cloud environment and controlled for smooth load distribution. Five different parameters, such as pitch frequency (PF), standard pitch delay (SPD), average distance between peaks (ADP), note duration gradient (NDG) and pitch class gradient (PCG), are leveraged to assess the quality of the improvised music. The proposed model obtains +0.15 of PF, −0.43 of SPD, −0.07 of ADP and 0.0041 NDG, which is a better value than other improvisation methods.

Background Brightness-mode (B-mode) ultrasound is a valuable tool to non-invasively image skeletal muscle architectural changes during movement, but automatically tracking muscle fascicles remains a major challenge. Existing fascicle tracking algorithms either require time-consuming drift corrections or yield noisy estimates that require post-processing. We therefore aimed to develop an algorithm that tracks fascicles without drift and with low noise across a range of experimental conditions and image acquisition settings. Methods We applied a Kalman filter to combine fascicle length and fascicle angle estimates from existing and openly-available UltraTrack and TimTrack algorithms into a hybrid algorithm called UltraTimTrack. We applied the hybrid algorithm to ultrasound image sequences collected from the human medial gastrocnemius of healthy individuals ( N = 8, four women), who performed cyclical submaximal plantar flexion contractions or remained at rest during passive ankle joint rotations at given frequencies and amplitudes whilst seated in a dynamometer chair. We quantified the algorithm’s tracking accuracy, noise, and drift as the respective mean, cycle-to-cycle variability, and accumulated between-contraction variability in fascicle length and fascicle angle. We expected UltraTimTrack’s estimates to be less noisy than TimTrack’s estimates and to drift less than UltraTrack’s estimates across a range of conditions and image acquisition settings. Results The proposed algorithm yielded low-noise estimates like UltraTrack and was drift-free like TimTrack across the broad range of conditions we tested. Over 120 cyclical contractions, fascicle length and fascicle angle deviations of UltraTimTrack accumulated to 2.1 ± 1.3 mm (mean ± sd) and 0.8 ± 0.7 deg, respectively. This was considerably less than UltraTrack (67.0 ± 59.3 mm, 9.3 ± 8.6 deg) and similar to TimTrack (1.9 ± 2.2 mm, 0.9 ± 1.0 deg). Average cycle-to-cycle variability of UltraTimTrack was 1.4 ± 0.4 mm and 0.6 ± 0.3 deg, which was similar to UltraTrack (1.1 ± 0.3 mm, 0.5 ± 0.1 deg) and less than TimTrack (3.5 ± 1.0 mm, 1.4 ± 0.5 deg). UltraTimTrack was less affected by experimental conditions and image acquisition settings than its parent algorithms. It also yielded similar or lower root-mean-square deviations from manual tracking for previously published image sequences (fascicle length: 2.3–2.6 mm, fascicle angle: 0.8–0.9 deg) compared with a recently-proposed hybrid algorithm (4.7 mm, 0.9 deg), and the recently-proposed DL_Track algorithm (3.8 mm, 3.9 deg). Furthermore, UltraTimTrack’s processing time (0.2 s per image) was at least five times shorter than that of these recently-proposed algorithms. Conclusion We developed a Kalman-filter-based algorithm to improve fascicle tracking from B-mode ultrasound image sequences. The proposed algorithm provides low-noise, drift-free estimates of muscle architectural changes that may better inform muscle function interpretations.

The k nearest neighbor (kNN) approach is a simple and effective algorithm for classification and a number of variants have been proposed based on the kNN algorithm. One of the limitations of kNN is that the method may be less effective when data contains many noisy features due to their non-informative influence in calculating distance. Additionally, information derived from nearest neighbors may be less meaningful in high-dimensional data. To address the limitation of nearest-neighbor based approaches in high-dimensional data, we propose to extend the k conditional nearest neighbor (kCNN) method which is an effective variant of kNN. The proposed approach aggregates multiple kCNN classifiers, each constructed from a randomly sampled feature subset. We also develop a score metric to weigh individual classifiers based on the level of separation of the feature subsets. We investigate the properties of the proposed method using simulation. Moreover, the experiments on gene expression datasets show that the proposed method is promising in terms of predictive classification performance.

Patient privacy data security is a pivotal area of research within the burgeoning field of smart healthcare. This study proposes an innovative hybrid blockchain-based framework for the secure sharing of electronic medical record (EMR) data. Unlike traditional privacy protection schemes, our approach employs a novel tripartite blockchain architecture that segregates healthcare data across distinct blockchains for patients and healthcare providers while introducing a separate social blockchain to enable privacy-preserving data sharing with authorized external entities. This structure enhances both security and transparency while fostering collaborative efforts across different stakeholders. To address the inherent complexity of managing multiple blockchains, a unique cross-chain signature algorithm is introduced, based on the Boneh-Lynn-Shacham (BLS) signature aggregation technique. This algorithm not only streamlines the signature process across chains but also strengthens system security and optimizes storage efficiency, addressing a key challenge in multi-chain systems. Additionally, our external sharing algorithm resolves the prevalent issue of medical data silos by facilitating better data categorization and enabling selective, secure external sharing through the social blockchain. Security analyses and experimental results demonstrate that the proposed scheme offers superior security, storage optimization, and flexibility compared to existing solutions, making it a robust choice for safeguarding patient data in smart healthcare environments.

Temporal knowledge graphs (TKGs) are critical tools for capturing the dynamic nature of facts that evolve over time, making them highly valuable in a broad spectrum of intelligent applications. In the domain of temporal knowledge graph extrapolation reasoning, the prediction of future occurrences is of great significance and presents considerable obstacles. While current models consider the fact changes over time and recognize that historical facts may recur, they often overlook the influence of past events on future predictions. Motivated by these considerations, this work introduces a novel temporal knowledge graph reasoning model, named Temporal Reasoning with Recurrent Encoding and Contrastive Learning (TRCL), which integrates recurrent encoding and contrastive learning techniques. The proposed model has the ability to capture the evolution of historical facts, generating representations of entities and relationships through recurrent encoding. Additionally, TRCL incorporates a global historical matrix to account for repeated historical occurrences and employs contrastive learning to alleviate the interference of historical facts in predicting future events. The TKG reasoning outcomes are subsequently derived through a time decoder. A quantity of experiments conducted on four benchmark datasets demonstrate the exceptional performance of the proposed TRCL model across a range of metrics, surpassing state-of-the-art TKG reasoning models. When compared to the strong baseline Time-Guided Recurrent Graph Network (TiRGN) model, the proposed TRCL achieves 1.03% improvements on ICEWS14 using mean reciprocal rank (MRR) evaluation metric. This innovative proposed method not only enhances the accuracy of TKG extrapolation, but also sets a new standard for robustness in dynamic knowledge graph applications, paving the way for future research and practical applications in predictive intelligence systems.

Scene graph generation (SGG) aims to identify and extract objects from images and elucidate their interrelations. This task faces two primary challenges. Firstly, the long-tail distribution of relation categories causes SGG models to favor high-frequency relations, such as “ on” and “ in” . Secondly, some subject-object pairs may have multiple reasonable relations, which often possess a certain degree of semantic similarity. However, the use of one-hot ground-truth relation labels does not effectively represent the semantic similarities and distinctions among relations. In response to these challenges, we propose a model-agnostic method named Mixup and Balanced Relation Learning (MBRL). This method assigns soft labels to samples exhibiting semantic ambiguities and optimizes model training by adjusting the loss weights for fine-grained and low-frequency relation samples. Its model-agnostic design facilitates seamless integration with diverse SGG models, enhancing their performance across various relation categories. Our approach is evaluated on widely-used datasets, including Visual Genome and Generalized Question Answering, both with over 100,000 images, providing rich visual contexts for scene graph model evaluation. Experimental results show that our method outperforms state-of-the-art approaches on multiple scene graph generation tasks, demonstrating significant improvements in both relation prediction accuracy and the handling of imbalanced data distributions.

The inconsistency in software development standards frequently leads to vulnerabilities that can jeopardize an application’s cryptographic integrity. This situation can result in incomplete or flawed encryption processes. Vulnerabilities may manifest as missing, bypassed, or improperly executed encryption functions or the absence of critical cryptographic mechanisms, which eventually weaken security goals. This article introduces a thorough method for detecting vulnerabilities using dynamic and static analysis, focusing on a cryptographic function dominance tree. This strategy systematically minimizes the likelihood of integrity breaches in cryptographic applications. A layered and modular model is developed to maintain integrity by mapping the entire flow of cryptographic function calls across various components. The cryptographic function call graph and dominance tree are extracted and subsequently analyzed using an integrated dynamic and static technique. The extracted information undergoes strict evaluation against the anticipated function call sequence in the relevant cryptographic module to identify and localize potential security issues. Experimental findings demonstrate that the proposed method considerably enhances the accuracy and comprehensiveness of vulnerability detection in cryptographic applications, improving implementation security and resilience against misuse vulnerabilities.

Graph auto-encoders are a crucial research area within graph neural networks, commonly employed for generating graph embeddings while minimizing errors in unsupervised learning. Traditional graph auto-encoders focus on reconstructing minimal graph data loss to encode neighborhood information for each node, yielding node embedding representations. However, existing graph auto-encoder models often overlook node representations and fail to capture contextual node information within the graph data, resulting in poor embedding effects. Accordingly, this study proposes the ensemble graph auto-encoders (E-GAE) model. It utilizes the ensemble random walk graph auto-encoder, the random walk graph auto-encoder of the ensemble network, and the graph attention auto-encoder to generate three node embedding matrices Z. Then, these techniques are combined using adaptive weights to reconstruct a new node embedding matrix. This method addresses the problem of low-quality embeddings. The model’s performance is evaluated using three publicly available datasets (Cora, Citeseer, and PubMed), indicating its effectiveness through multiple experiments. It achieves up to a 2.0% improvement in the link prediction task and a 9.4% enhancement in the clustering task. Our code for this work can be found at https://github.com/xcgydfjjjderg/graphautoencoder .

News background linking is the problem of finding useful links to resources that provide contextual background information for a given news article. Many systems were proposed to address this problem. Yet, the most effective and reproducible method, to date, used the entire input article as a search query to retrieve the background links by sparse retrieval. While being effective, that method is still far from being optimal. Furthermore, it only leverages the lexical matching signal between the input article and the candidate background links. Nevertheless, intuitively, there may exist resources with useful background information that do not lexically overlap with the input article’s vocabulary. While many studies proposed systems that adopt semantic matching for addressing news background linking, none were able to outperform the simple lexical-based matching method. In this paper, we investigate multiple methods to integrate both the lexical and semantic relevance signals for better reranking of candidate background links. To represent news articles in the semantic space, we compare multiple Transformer-based encoder models in a zero-shot setting without the need for any labeled data. Our results show that using a hierarchical aggregation of sentence-level representations generates a good semantic representation of news articles, which is then integrated with lexical matching to achieve a new state-of-the-art solution for the problem. We further show that a significant performance improvement is potentially attainable if the degree by which a semantic relevance signal is needed is accurately predicted per input article.

Road damage often includes potholes, cracks, lane degradation, and surface shading. Potholes are a common problem in pavements. Detecting them is crucial for maintaining infrastructure and ensuring public safety. A thorough assessment of pavement conditions is required before planning any preventive repairs. Herein, we report the use of transfer learning and deep learning (DL) models to preprocess digital images of pavements for better pothole detection. Fourteen models were evaluated, including MobileNet, MobileNetV2, NASNetMobile, DenseNet121, DenseNet169, InceptionV3, DenseNet201, ResNet152V2, EfficientNetB0, InceptionResNetV2, Xception, and EfficientNetV2M. The study introduces a modified MobileNetV2 (MMNV2) model designed for fast and efficient feature extraction. The MMNV2 model exhibits improved classification, detection, and prediction accuracy by adding a five-layer pre-trained network to the MobileNetV2 framework. It combines deep learning, deep neural networks (DNN), and transfer learning, which resulted in better performance compared to other models. The MMNV2 model was tested using a dataset of 5,000 pavement images. A learning rate of 0.001 was used to optimize the model. It classified images into ‘normal’ or ‘pothole’ categories with 99.95% accuracy. The model also achieved 100% recall, 99.90% precision, 99.95% F1-score, and a 0.05% error rate. The MMNV2 model uses fewer parameters while delivering better results. It offers a promising solution for real-world applications in pothole detection and pavement assessment.

One of the most complex and life-threatening pathologies of the central nervous system is brain tumors. Correct diagnosis of these tumors plays an important role in determining the treatment plans of patients. Traditional classification methods often rely on manual assessments, which can be prone to error. Therefore, multiple classification of brain tumors has gained significant interest in recent years in both the medical and computer science fields. The use of artificial intelligence and machine learning, especially in the automatic classification of brain tumors, is increasing significantly. Deep learning models can achieve high accuracy when trained on datasets in diagnosis and classification. This study examined deep learning-based approaches for automatic multi-class classification of brain tumors, and a new approach combining deep learning and quantum genetic algorithms (QGA) was proposed. The powerful feature extraction ability of the pre-trained EfficientNetB0 was utilized and combined with this quantum genetic algorithms, a new approach was proposed. It is aimed to develop the feature selection method. With this hybrid method, high reliability and accuracy in brain tumor classification was achieved. The proposed model achieved high accuracy of 98.36% and 98.25%, respectively, with different data sets and significantly outperformed traditional methods. As a result, the proposed method offers a robust and scalable solution that will help classify brain tumors in early and accurate diagnosis and contribute to the field of medical imaging with patient outcomes.

Background Foreign object debris (FOD) is an unwanted substance that damages vehicular systems, most commonly the wheels of vehicles. In airport runways, these foreign objects can damage the wheels or internal systems of planes, potentially leading to flight crashes. Surveys indicate that FOD-related damage costs over $4 billion annually, affecting airlines, airport tenants, and passengers. Current FOD clearance involves high-cost radars and significant manpower, and existing radar and camera-based surveillance methods are expensive to install. Methods This work proposes a video-based deep learning methodology to address the high cost of radar-based FOD detection. The proposed system consists of two modules for FOD detection: object classification and object localization. The classification module categorizes FOD into specific types of foreign objects. In the object localization module, these classified objects are pinpointed in video frames. Results The proposed system was experimentally tested with a large video dataset and compared with existing methods. The results demonstrated improved accuracy and robustness, allowing the FOD clearance team to quickly detect and remove foreign objects, thereby enhancing the safety and efficiency of airport runway operations.

The ejection fraction (EF) is a vital metric for assessing cardiovascular function through cardiac ultrasound. Manual evaluation is time-consuming and exhibits high variability among observers. Deep-learning techniques offer precise and autonomous EF predictions, yet these methods often lack explainability. Accurate heart failure prediction using cardiac ultrasound is challenging due to operator dependency and inconsistent video quality, resulting in significant interobserver variability. To address this, we developed a method integrating convolutional neural networks (CNN) and transformer models for direct EF estimation from ultrasound video scans. This article introduces a Residual Transformer Module (RTM) that extends a 3D ResNet-based network to analyze (2D + t) spatiotemporal cardiac ultrasound video scans. The proposed method, EFNet, utilizes cardiac ultrasound video images for end-to-end EF value prediction. Performance evaluation on the EchoNet-Dynamic dataset yielded a mean absolute error (MAE) of 3.7 and an R2 score of 0.82. Experimental results demonstrate that EFNet outperforms state-of-the-art techniques, providing accurate EF predictions.

The use of brain-computer interface (BCI) technology to identify emotional states has gained significant interest, especially with the rise of virtual reality (VR) applications. However, the extensive calibration required for precise emotion recognition models presents a significant challenge, particularly for sensitive groups such as children, elderly, and patients. This study presents a novel approach that utilizes heterogeneous adversarial transfer learning (HATL) to synthesize electroencephalography (EEG) data from various other signal modalities, reducing the need for lengthy calibration phases. We benchmark the efficacy of three generative adversarial network (GAN) architectures, such as conditional GAN (CGAN), conditional Wasserstein GAN (CWGAN), and CWGAN with gradient penalty (CWGAN-GP) within this framework. The proposed framework is rigorously tested on two conventional open sourced datasets, SEED-V and DEAP. Additionally, the framework was applied to an immersive three-dimensional (3D) dataset named GraffitiVR, which we collected to capture the emotional and behavioral reactions of individuals experiencing urban graffiti in a VR environment. This expanded application provides insights into emotion recognition frameworks in VR settings, providing a wider range of contexts for assessing our methodology. When the accuracy of emotion recognition classifiers trained with CWGAN-GP-generated EEG data combined with non-EEG sensory data was compared against those trained using a combination of real EEG and non-EEG sensory data, the accuracy ratios were 93% on the SEED-V dataset, 99% on the DEAP dataset, and 97% on the GraffitiVR dataset. Moreover, in the GraffitiVR dataset, using CWGAN-GP-generated EEG data with non-EEG sensory data for emotion recognition models resulted in up to a 30% reduction in calibration time compared to classifiers trained on real EEG data with non-EEG sensory data. These results underscore the robustness and versatility of the proposed approach, significantly enhancing emotion recognition processes across a variety of environmental settings.

Diabetes is a disease that affects millions of people in the world and its early screening prevents serious health problems, also providing relief in the demand for healthcare services. In the search for methods to support early diagnosis, this article introduces a novel prediabetes risk classification algorithm (PRCA) for type-2 diabetes mellitus (T2DM), utilizing the chemosensitivity of carotid bodies (CB) and K-means clustering technique from the field of machine learning. Heart rate (HR) and respiratory rate (RR) data from eight volunteers with prediabetes and 25 without prediabetes were analyzed. Data were collected in basal conditions and after stimulation of the CBs by inhalation of 100% of oxygen and after ingestion of a standardized meal. During the analysis, a greater variability of groups was observed in people with prediabetes compared to the control group, particularly after inhalation of oxygen. The algorithm developed from these results showed an accuracy of 86% in classifying for prediabetes. This approach, centered on CB chemosensitivity deregulation in early disease stages, offers a nuanced detection method beyond conventional techniques. Moreover, the adaptable algorithm and clustering methodology hold promise as risk classifications for other diseases. Future endeavors aim to validate the algorithm through longitudinal studies tracking disease development among volunteers and expand the study’s scope to include a larger participant pool.

Planning personalized travel itineraries for groups with diverse preferences is indeed challenging. This article proposes a novel group tour trip recommender model (GTTRM), which uses ant colony optimization (ACO) to optimize group satisfaction while minimizing conflicts between group members. Unlike existing models, the proposed GTTRM allows dynamic subgroup formation during the trip to handle conflicting preferences and provide tailored recommendations. Experimental results show that GTTRM significantly improves satisfaction levels for individual group members, outperforming state-of-the-art models in terms of both subgroup management and optimization efficiency.

Binary code similarity detection (BCSD) aims to identify whether a pair of binary code snippets is similar, which is widely used for tasks such as malware analysis, patch analysis, and clone detection. Current state-of-the-art approaches are based on Transformer, which require substantial computation resources. Learning-based approaches remains room for optimization in learning the deeper semantics of binary code. In this paper, we propose MSSA, a multi-stage semantic-aware neural network for BCSD at the function level. It effectively integrates the semantic and structural information of assembly instructions within and between basic blocks, and across the entire function through four semantic-aware neural networks, achieving deep understanding of binary code semantics. MSSA is a lightweight model with only 0.38M parameters in its backbone network, suitable for deployment in CPU environments. Experimental results show that MSSA outperforms Gemini, Asm2Vec, SAFE, and jTrans in classification performance and ranks second only to the Transformer-based jTrans in retrieval performance.

Technological advancements have significantly enhanced e-commerce, helping customers find the best products. One key development is recommendation systems, which personalize the shopping experience and boost sales. This paper explores a novel geographic recommendation system that uses demographic data, such as population density, age, and income, to refine recommendations. By integrating geographic and demographic information, like the population size of a country, businesses can tailor their offerings to regional preferences. This targeted approach aims to make recommendations more relevant by considering the behaviors and needs of different geographic areas. We sourced population data from The National Institute of Statistics (Tunisia, INS). This approach improves the importance of product recommendations for particular locations by customizing them based on demographic and geographic measures. The technique creates a better context-aware recommendation system that boosts customer happiness and business proceeds by fusing consumer behavior with extensive demographic data. The method also includes a mathematical model that considers population intensity to refine further recommendations established on the regional model.

Image denoising is a complex task that always yields an approximated version of the clean image. Unfortunately, the existing works have focussed only on the peak signal to noise ratio (PSNR) metric and have shown no attention to edge features in a reconstructed image. Although fully convolution neural networks (CNN) are capable of removing the noise using kernel filters and automatic extraction of features, it has failed to reconstruct the images for higher values of noise standard deviation. Additionally, deep learning models require a huge database to learn better from the inputs. This, in turn, increases the computational complexity and memory requirement. Therefore, we propose the Median Noise Residue U-Net (MNRU-Net) with a limited training database without involving image augmentation. In the proposed work, the learning capability of the traditional U-Net model was increased by adding hand-crafted features in the input layers of the U-Net. Further, an approximate version of the noise estimated from the median filter and the gradient information of the image were used to improve the performance of U-Net. Later, the performance of MNRU-Net was evaluated based on PSNR, structural similarity, and figure of merit for different noise standard deviations of 15, 25, and 50 respectively. It is witnessed that the results gained from the suggested work are better than the results yielded by complex denoising models such as the robust deformed denoising CNN (RDDCNN). This work emphasizes that the skip connections along with the hand-crafted features could improve the performance at higher noise levels by using this simple architecture. In addition, the model was found to be less expensive, with low computational complexity.

With the escalating number of vehicles and the lack of parking spaces, the issue of parking has become a significant problem in major cities as it is a daily occurrence for educational institutions, companies, and government facilities, resulting in fuel wastage and time inefficiencies. In their work lives, employees often face problems when parking their cars in the work parking area. Finding a space for their vehicle can take a lot of time and effort, leading to late arrival for work. On the other hand, security guards have difficulty entering their employees’ cars. In this context, our proposed system attempts to address this pressing issue, which consists of two parts: one is a camera at the parking gate that recognizes the license plate using the Automatic Number Plate Recognition (ANPR) algorithm, where the camera captures the license plate and outputs the plate number using the optical character recognition (OCR) technique. After that, the resulting data is cross-referenced with database records for seamless entry authentication. This eliminates the need for security personnel to verify vehicle identities or stickers manually, streamlining access procedures. The second part is a camera in the car parks that distinguishes between vacant and available parking spaces and stores the data collected by the camera in the centralized database, enabling the real-time display of the nearest available parking spots on digital screens at entrance gates, significantly reducing the time and effort spent in locating parking spaces. Through this innovative solution, we aim to enhance urban mobility and alleviate the challenges associated with urban parking congestion, thereby resolving the problem of intelligent parking for smart cities with the help of machine learning.

Blink detection is a highly concerned research direction in the field of computer vision, which plays a key role in various application scenes such as human-computer interaction, fatigue detection and emotion perception. In recent years, with the rapid development of deep learning, the application of deep learning techniques for precise blink detection has emerged as a significant area of interest among researchers. Compared with traditional methods, the blink detection method based on deep learning offers superior feature learning ability and higher detection accuracy. However, the current research on blink detection based on deep learning lacks systematic summarization and comparison. Therefore, the aim of this article is to comprehensively review the research progress in deep learning-based blink detection methods and help researchers to have a clear understanding of the various approaches in this field. This article analyzes the progress made by several classical deep learning models in practical applications of eye blink detection while highlighting their respective strengths and weaknesses. Furthermore, it provides a comprehensive summary of commonly used datasets and evaluation metrics for blink detection. Finally, it discusses the challenges and future directions of deep learning for blink detection applications. Our analysis reveals that deep learning-based blink detection methods demonstrate strong performance in detection. However, they encounter several challenges, including training data imbalance, complex environment interference, real-time processing issues and application device limitations. By overcoming the challenges identified in this study, the application prospects of deep learning-based blink detection algorithms will be significantly enhanced.

Republicans and Democrats practically everywhere have been demonstrating concerns about environmental conservation to achieve sustainable development goals (SDGs) since the turn of the century. To promote fuel (energy) savings and a reduction in the amount of carbon dioxide CO 2 emissions in several enterprises, actions have been taken based on the concepts described. This study proposes an environmentally friendly manufacturing system designed to minimize environmental impacts. Specifically, it aims to develop a sustainable manufacturing process that accounts for energy consumption and CO 2 emissions from direct and indirect energy sources. A multi-objective mathematical model has been formulated, incorporating financial and environmental constraints, to minimize overall costs, energy consumption, and CO 2 emissions within the manufacturing framework. The input model parameters for real-world situations are generally unpredictable, so a fuzzy multi-objective model will be developed as a way to handle it. The validity of the proposed ecological industrial design will be tested using a scenario-based approach. Results demonstrate the high reliability, applicability, and effectiveness of the proposed network when analyzed using the developed techniques.

In the ever-expanding digital landscape, the abundance of user-generated content on consumer platforms such as Booking and TripAdvisor offers a rich source of information for both travellers and hoteliers. Sentiment analysis, a fundamental research task of natural language processing (NLP) is used for mining sentiments and opinions within this vast reservoir of text reviews. A more specific type of sentiment analysis, i.e ., aspect-based sentiment analysis (ABSA), is used when processing customer reviews is required. In ABSA, we aim to capture aspect-level sentiments and intricate relationships between various aspects within reviews. This article proposes a novel approach to ABSA by introducing a novel technique of word sense disambiguation (WSD) and integrating it with the Transformer architecture bidirectional encoder representations from Transformers (BERT) and graph convolutional networks (GCNs). The proposed approach resolves the intriguing ambiguities of the words and represents the review data as a complex graph structure, facilitating the modeling of intricate relationships between different aspects. The combination of bidirectional long short-term memory (BiLSTM) and GCN proves effective in capturing inter-dependencies among various aspects, providing a nuanced understanding of customer sentiments. The experiments are conducted on the RABSA dataset (an enhanced and richer hotel review data collection), and results demonstrate that our approach outperforms previous baselines, showcasing the effectiveness of integrating WSD in ABSA. Furthermore, an ablation study confirms the significant contribution of the WSD module to the overall performance. Moreover, we explore different similarity measures and find that cosine similarity yields the best results when identifying the real sense of a word in a given sentence using WordNet. The findings of our work and future work related to our work create lots of interest for people in the tourism and hospitality industry. This research gives another boost to the concept of the potential of NLP techniques in sentiment analysis. It emphasizes that if we combine the potential of NLP techniques along with state-of-the-art machine learning frameworks, we can shape the future of this field.

In intertidal mudflat culture (IMC), the fishing efficiency and the degree of damage to nature have always been a pair of irreconcilable contradictions. To improve the efficiency of razor clam fishing and at the same time reduce the damage to the natural environment, in this study, a razor clam burrows dataset is established, and an intelligent razor clam fishing method is proposed, which realizes the accurate identification and counting of razor clam burrows by introducing the object detection technology into the razor clam fishing activity. A detection model called intertidal mudflat culture-You Only Look Once (IMC-YOLO) is proposed in this study by making improvements upon You Only Look Once version 8 (YOLOv8). In this study, firstly, at the end of the backbone network, the Iterative Attention-based Intrascale Feature Interaction (IAIFI) module module was designed and adopted to improve the model’s focus on advanced features. Subsequently, to improve the model’s effectiveness in detecting difficult targets such as razor clam burrows with small sizes, the head network was refactored. Then, FasterNet Block is used to replace the Bottleneck, which achieves more effective feature extraction while balancing detection accuracy and model size. Finally, the Three Branch Convolution Attention Mechanism (TBCAM) is proposed, which enables the model to focus on the specific region of interest more accurately. After testing, IMC-YOLO achieved mAP50, mAP50:95, and F1best of 0.963, 0.636, and 0.918, respectively, representing improvements of 2.2%, 3.5%, and 2.4% over the baseline model. Comparison with other mainstream object detection models confirmed that IMC-YOLO strikes a good balance between accuracy and numbers of parameters.