Computers, Materials & Continua

Real-time detection of unhealthy fish remains a significant challenge in intensive recirculating aquaculture. Early recognition of unhealthy fish and the implementation of appropriate treatment measures are crucial for preventing the spread of diseases and minimizing economic losses. To address this issue, an improved algorithm based on the You Only Look Once v5s (YOLOv5s) lightweight model has been proposed. This enhanced model incorporates a faster lightweight structure and a new Convolutional Block Attention Module (CBAM) to achieve high recognition accuracy. Furthermore, the model introduces the α-SIoU loss function, which combines the α-Intersection over Union (α-IoU) and Shape Intersection over Union (SIoU) loss functions, thereby improving the accuracy of bounding box regression and object recognition. The average precision of the improved model reaches 94.2% for detecting unhealthy fish, representing increases of 11.3%, 9.9%, 9.7%, 2.5%, and 2.1% compared to YOLOv3-tiny, YOLOv4, YOLOv5s, GhostNet-YOLOv5, and YOLOv7, respectively. Additionally, the improved model positively impacts hardware efficiency, reducing requirements for memory size by 59.0%, 67.0%, 63.0%, 44.7%, and 55.6% in comparison to the five models mentioned above. The experimental results underscore the effectiveness of these approaches in addressing the challenges associated with fish health detection, and highlighting their significant practical implications and broad application prospects.

The design of the loading path is one of the important research contents of the tube hydroforming process. Optimization of loading paths using optimization algorithms has received attention due to the inefficiency of only finite element optimization. In this paper, the hydroforming process of 5A02 aluminum alloy variable diameter tube was as the research object. Fuzzy control was used to optimize the loading path, and the fuzzy rule base was established based on FEM. The minimum wall thickness and wall thickness reduction rate were determined as input membership functions, and the axial feeds variable value of the next step was used as output membership functions. The results show that the optimized loading path greatly improves the uniformity of wall thickness and the forming effect compared with the linear loading path. The round corner lamination rate of the tube is 91.2% under the fuzzy control optimized loading path, which was increased by 47.1% and 22.6% compared with linear loading Path 1 and Path 2, respectively. Based on the optimized loading path in the experiment, the minimum wall thickness of the variable diameter tube was 1.32 mm and the maximum thinning rate was 12.4%. The experimental results were consistent with the simulation results, which verified the accuracy of fuzzy control. The research results provide a reference for improving the forming quality of thin-walled tubes and plates.

To address the challenges of video copyright protection and ensure the perfect recovery of original video, we propose a dual-domain watermarking scheme for digital video, inspired by Robust Reversible Watermarking (RRW) technology used in digital images. Our approach introduces a parameter optimization strategy that incrementally adjusts scheme parameters through attack simulation fitting, allowing for adaptive tuning of experimental parameters. In this scheme, the low-frequency Polar Harmonic Transform (PHT) moment is utilized as the embedding domain for robust watermarking, enhancing stability against simulation attacks while implementing the parameter optimization strategy. Through extensive attack simulations across various digital videos, we identify the optimal low-frequency PHT moment using adaptive normalization. Subsequently, the embedding parameters for robust watermarking are adaptively adjusted to maximize robustness. To address computational efficiency and practical requirements, the unnormalized high-frequency PHT moment is selected as the embedding domain for reversible watermarking. We optimize the traditional single-stage extended transform dithering modulation (STDM) to facilitate multi-stage embedding in the dual-domain watermarking process. In practice, the video embedded with a robust watermark serves as the candidate video. This candidate video undergoes simulation according to the parameter optimization strategy to balance robustness and embedding capacity, with adaptive determination of embedding strength. The reversible watermarking is formed by combining errors and other information, utilizing recursive coding technology to ensure reversibility without attacks. Comprehensive analyses of multiple performance indicators demonstrate that our scheme exhibits strong robustness against Common Signal Processing (CSP) and Geometric Deformation (GD) attacks, outperforming other advanced video watermarking algorithms under similar conditions of invisibility, reversibility, and embedding capacity. This underscores the effectiveness and feasibility of our attack simulation fitting strategy.

The traffic within data centers exhibits bursts and unpredictable patterns. This rapid growth in network traffic has two consequences: it surpasses the inherent capacity of the network’s link bandwidth and creates an imbalanced network load. Consequently, persistent overload situations eventually result in network congestion. The Software Defined Network (SDN) technology is employed in data centers as a network architecture to enhance performance. This paper introduces an adaptive congestion control strategy, named DA-DCTCP, for SDN-based Data Centers. It incorporates Explicit Congestion Notification (ECN) and Round-Trip Time (RTT) to establish congestion awareness and an ECN marking model. To mitigate incorrect congestion caused by abrupt flows, an appropriate ECN marking is selected based on the queue length and its growth slope, and the congestion window (CWND) is adjusted by calculating RTT. Simultaneously, the marking threshold for queue length is continuously adapted using the current queue length of the switch as a parameter to accommodate changes in data centers. The evaluation conducted through Mininet simulations demonstrates that DA-DCTCP yields advantages in terms of throughput, flow completion time (FCT), latency, and resistance against packet loss. These benefits contribute to reducing data center congestion, enhancing the stability of data transmission, and improving throughput.

Object detection has made a significant leap forward in recent years. However, the detection of small objects continues to be a great difficulty for various reasons, such as they have a very small size and they are susceptible to missed detection due to background noise. Additionally, small object information is affected due to the downsampling operations. Deep learning-based detection methods have been utilized to address the challenge posed by small objects. In this work, we propose a novel method, the Multi-Convolutional Block Attention Network (MCBAN), to increase the detection accuracy of minute objects aiming to overcome the challenge of information loss during the downsampling process. The multi-convolutional attention block (MCAB); channel attention and spatial attention module (SAM) that make up MCAB, have been crafted to accomplish small object detection with higher precision. We have carried out the experiments on the Karlsruhe Institute of Technology and Toyota Technological Institute (KITTI) and Pattern Analysis, Statical Modeling and Computational Learning (PASCAL) Visual Object Classes (VOC) datasets and have followed a step-wise process to analyze the results. These experiment results demonstrate that significant gains in performance are achieved, such as 97.75% for KITTI and 88.97% for PASCAL VOC. The findings of this study assert quite unequivocally the fact that MCBAN is much more efficient in the small object detection domain as compared to other existing approaches.

Electricity is essential for keeping power networks balanced between supply and demand, especially since it costs a lot to store. The article talks about different deep learning methods that are used to guess how much green energy different Asian countries will produce. The main goal is to make reliable and accurate predictions that can help with the planning of new power plants to meet rising demand. There is a new deep learning model called the Green-electrical Production Ensemble (GP-Ensemble). It combines three types of neural networks: convolutional neural networks (CNNs), gated recurrent units (GRUs), and feedforward neural networks (FNNs). The model promises to improve prediction accuracy. The 1965–2023 dataset covers green energy generation statistics from ten Asian countries. Due to the rising energy supply-demand mismatch, the primary goal is to develop the best model for predicting future power production. The GP-Ensemble deep learning model outperforms individual models (GRU, FNN, and CNN) and alternative approaches such as fully convolutional networks (FCN) and other ensemble models in mean squared error (MSE), mean absolute error (MAE) and root mean squared error (RMSE) metrics. This study enhances our ability to predict green electricity production over time, with MSE of 0.0631, MAE of 0.1754, and RMSE of 0.2383. It may influence laws and enhance energy management.

In Saudi Arabia, drones are increasingly used in different sensitive domains like military, health, and agriculture to name a few. Typically, drone cameras capture aerial images of objects and convert them into crucial data, alongside collecting data from distributed sensors supplemented by location data. The interception of the data sent from the drone to the station can lead to substantial threats. To address this issue, highly confidential protection methods must be employed. This paper introduces a novel steganography approach called the Shuffling Steganography Approach (SSA). SSA encompasses five fundamental stages and three proposed algorithms, designed to enhance security through strategic encryption and data hiding techniques. Notably, this method introduces advanced resistance to brute force attacks by employing predefined patterns across a wide array of images, complicating unauthorized access. The initial stage involves encryption, dividing, and disassembling the encrypted data. A small portion of the encrypted data is concealed within the text (Algorithm 1) in the third stage. Subsequently, the parts are merged and mixed (Algorithm 2), and finally, the composed text is hidden within an image (Algorithm 3). Through meticulous investigation and comparative analysis with existing methodologies, the proposed approach demonstrates superiority across various pertinent criteria, including robustness, secret message size capacity, resistance to multiple attacks, and multilingual support.

In video surveillance, anomaly detection requires training machine learning models on spatio-temporal video sequences. However, sometimes the video-only data is not sufficient to accurately detect all the abnormal activities. Therefore, we propose a novel audio-visual spatiotemporal autoencoder specifically designed to detect anomalies for video surveillance by utilizing audio data along with video data. This paper presents a competitive approach to a multi-modal recurrent neural network for anomaly detection that combines separate spatial and temporal autoencoders to leverage both spatial and temporal features in audio-visual data. The proposed model is trained to produce low reconstruction error for normal data and high error for abnormal data, effectively distinguishing between the two and assigning an anomaly score. Training is conducted on normal datasets, while testing is performed on both normal and anomalous datasets. The anomaly scores from the models are combined using a late fusion technique, and a deep dense layer model is trained to produce decisive scores indicating whether a sequence is normal or anomalous. The model’s performance is evaluated on the University of California, San Diego Pedestrian 2 (UCSD PED 2), University of Minnesota (UMN), and Tampere University of Technology (TUT) Rare Sound Events datasets using six evaluation metrics. It is compared with state-of-the-art methods depicting a high Area Under Curve (AUC) and a low Equal Error Rate (EER), achieving an (AUC) of 93.1 and an (EER) of 8.1 for the (UCSD) dataset, and an (AUC) of 94.9 and an (EER) of 5.9 for the UMN dataset. The evaluations demonstrate that the joint results from the combined audio-visual model outperform those from separate models, highlighting the competitive advantage of the proposed multi-modal approach.

In today’s world, image processing techniques play a crucial role in the prognosis and diagnosis of various diseases due to the development of several precise and accurate methods for medical images. Automated analysis of medical images is essential for doctors, as manual investigation often leads to inter-observer variability. This research aims to enhance healthcare by enabling the early detection of diabetic retinopathy through an efficient image processing framework. The proposed hybridized method combines Modified Inertia Weight Particle Swarm Optimization (MIWPSO) and Fuzzy C-Means clustering (FCM) algorithms. Traditional FCM does not incorporate spatial neighborhood features, making it highly sensitive to noise, which significantly affects segmentation output. Our method incorporates a modified FCM that includes spatial functions in the fuzzy membership matrix to eliminate noise. The results demonstrate that the proposed FCM-MIWPSO method achieves highly precise and accurate medical image segmentation. Furthermore, segmented images are classified as benign or malignant using the Decision Tree-Based Temporal Association Rule (DT-TAR) Algorithm. Comparative analysis with existing state-of-the-art models indicates that the proposed FCM-MIWPSO segmentation technique achieves a remarkable accuracy of 98.42% on the dataset, highlighting its significant impact on improving diagnostic capabilities in medical imaging.

Unsupervised learning methods such as graph contrastive learning have been used for dynamic graph representation learning to eliminate the dependence of labels. However, existing studies neglect positional information when learning discrete snapshots, resulting in insufficient network topology learning. At the same time, due to the lack of appropriate data augmentation methods, it is difficult to capture the evolving patterns of the network effectively. To address the above problems, a position-aware and subgraph enhanced dynamic graph contrastive learning method is proposed for discrete-time dynamic graphs. Firstly, the global snapshot is built based on the historical snapshots to express the stable pattern of the dynamic graph, and the random walk is used to obtain the position representation by learning the positional information of the nodes. Secondly, a new data augmentation method is carried out from the perspectives of short-term changes and long-term stable structures of dynamic graphs. Specifically, subgraph sampling based on snapshots and global snapshots is used to obtain two structural augmentation views, and node structures and evolving patterns are learned by combining graph neural network, gated recurrent unit, and attention mechanism. Finally, the quality of node representation is improved by combining the contrastive learning between different structural augmentation views and between the two representations of structure and position. Experimental results on four real datasets show that the performance of the proposed method is better than the existing unsupervised methods, and it is more competitive than the supervised learning method under a semi-supervised setting.

In the evolving landscape of the smart grid (SG), the integration of non-organic multiple access (NOMA) technology has emerged as a pivotal strategy for enhancing spectral efficiency and energy management. However, the open nature of wireless channels in SG raises significant concerns regarding the confidentiality of critical control messages, especially when broadcasted from a neighborhood gateway (NG) to smart meters (SMs). This paper introduces a novel approach based on reinforcement learning (RL) to fortify the performance of secrecy. Motivated by the need for efficient and effective training of the fully connected layers in the RL network, we employ an improved chimp optimization algorithm (IChOA) to update the parameters of the RL. By integrating the IChOA into the training process, the RL agent is expected to learn more robust policies faster and with better convergence properties compared to standard optimization algorithms. This can lead to improved performance in complex SG environments, where the agent must make decisions that enhance the security and efficiency of the network. We compared the performance of our proposed method (IChOA-RL) with several state-of-the-art machine learning (ML) algorithms, including recurrent neural network (RNN), long short-term memory (LSTM), K-nearest neighbors (KNN), support vector machine (SVM), improved crow search algorithm (I-CSA), and grey wolf optimizer (GWO). Extensive simulations demonstrate the efficacy of our approach compared to the related works, showcasing significant improvements in secrecy capacity rates under various network conditions. The proposed IChOA-RL exhibits superior performance compared to other algorithms in various aspects, including the scalability of the NOMA communication system, accuracy, coefficient of determination (R2), root mean square error (RMSE), and convergence trend. For our dataset, the IChOA-RL architecture achieved coefficient of determination of 95.77% and accuracy of 97.41% in validation dataset. This was accompanied by the lowest RMSE (0.95), indicating very precise predictions with minimal error.

In the process of dense vehicles traveling fast, there will be mutual occlusion between vehicles, which will lead to the problem of deterioration of the tracking effect of different vehicles, so this paper proposes a research method of virtual simulation video vehicle target tracking based on you only look once (YOLO)v5s and deep simple online and realtime tracking (DeepSort). Given that the DeepSort algorithm is currently the most effective tracking method, this paper merges the YOLOv5 algorithm with the DeepSort algorithm. Then it adds the efficient channel attention networks (ECA-Net) focusing mechanism at the back for the cross-stage partial bottleneck with 3 convolutions (C3) modules about the YOLOv5 backbone network and before the up-sampling of the Neck feature pyramid. The YOLOv5 algorithm adopts expected intersection over union (EIOU) instead of complete intersection over union (CIOU) as the loss function of the target frame regression. The improved YOLOv5 algorithm is named YOLOv5-Block. The experimental results show that in the special vehicle target detection (TD) and tracking in the virtual simulation environment, The YOLOv5-Block algorithm has an average accuracy (AP) of 99.5%, which significantly improves the target recognition correctness for typical occlusion cases, and is 1.48 times better than the baseline algorithm. After the virtual simulation video sequence test, multiple objects tracking accuracy (MOTA) and various objects tracking precision (MOTP) improved by 10.7 and 1.75 percentage points, respectively, and the number of vehicle target identity document (ID) switches decreased. Compared with recent mainstream vehicle detection and tracking models, the YOLOv5-Block+Deepsort algorithm can accurately and continuously complete the detection and tracking tasks of special vehicle targets in different scenes.

Infrared small target detection technology plays a pivotal role in critical military applications, including early warning systems and precision guidance for missiles and other defense mechanisms. Nevertheless, existing traditional methods face several significant challenges, including low background suppression ability, low detection rates, and high false alarm rates when identifying infrared small targets in complex environments. This paper proposes a novel infrared small target detection method based on a transformed Gaussian filter kernel and clustering approach. The method provides improved background suppression and detection accuracy compared to traditional techniques while maintaining simplicity and lower computational costs. In the first step, the infrared image is filtered by a new filter kernel and the results of filtering are normalized. In the second step, an adaptive thresholding method is utilized to determine the pixels in small targets. In the final step, a fuzzy C-mean clustering algorithm is employed to group pixels in the same target, thus yielding the detection results. The results obtained from various real infrared image datasets demonstrate the superiority of the proposed method over traditional approaches. Compared with the traditional method of state of the arts detection method, the detection accuracy of the four sequences is increased by 2.06%, 0.95%, 1.03%, and 1.01%, respectively, and the false alarm rate is reduced, thus providing a more effective and robust solution.

Ransomware has emerged as a critical cybersecurity threat, characterized by its ability to encrypt user data or lock devices, demanding ransom for their release. Traditional ransomware detection methods face limitations due to their assumption of similar data distributions between training and testing phases, rendering them less effective against evolving ransomware families. This paper introduces TLERAD (Transfer Learning for Enhanced Ransomware Attack Detection), a novel approach that leverages unsupervised transfer learning and co-clustering techniques to bridge the gap between source and target domains, enabling robust detection of both known and unknown ransomware variants. The proposed method achieves high detection accuracy, with an AUC of 0.98 for known ransomware and 0.93 for unknown ransomware, significantly outperforming baseline methods. Comprehensive experiments demonstrate TLERAD’s effectiveness in real-world scenarios, highlighting its adaptability to the rapidly evolving ransomware landscape. The paper also discusses future directions for enhancing TLERAD, including real-time adaptation, integration with lightweight and post-quantum cryptography, and the incorporation of explainable AI techniques.

Encrypted traffic plays a crucial role in safeguarding network security and user privacy. However, encrypting malicious traffic can lead to numerous security issues, making the effective classification of encrypted traffic essential. Existing methods for detecting encrypted traffic face two significant challenges. First, relying solely on the original byte information for classification fails to leverage the rich temporal relationships within network traffic. Second, machine learning and convolutional neural network methods lack sufficient network expression capabilities, hindering the full exploration of traffic’s potential characteristics. To address these limitations, this study introduces a traffic classification method that utilizes time relationships and a higher-order graph neural network, termed HGNN-ETC. This approach fully exploits the original byte information and chronological relationships of traffic packets, transforming traffic data into a graph structure to provide the model with more comprehensive context information. HGNN-ETC employs an innovative k-dimensional graph neural network to effectively capture the multi-scale structural features of traffic graphs, enabling more accurate classification. We select the ISCXVPN and the USTC-TK2016 dataset for our experiments. The results show that compared with other state-of-the-art methods, our method can obtain a better classification effect on different datasets, and the accuracy rate is about 97.00%. In addition, by analyzing the impact of varying input specifications on classification performance, we determine the optimal network data truncation strategy and confirm the model’s excellent generalization ability on different datasets.

Forecasting travel demand requires a grasp of individual decision-making behavior. However, transport mode choice (TMC) is determined by personal and contextual factors that vary from person to person. Numerous characteristics have a substantial impact on travel behavior (TB), which makes it important to take into account while studying transport options. Traditional statistical techniques frequently presume linear correlations, but real-world data rarely follows these presumptions, which may make it harder to grasp the complex interactions. Thorough systematic review was conducted to examine how machine learning (ML) approaches might successfully capture nonlinear correlations that conventional methods may ignore to overcome such challenges. An in-depth analysis of discrete choice models (DCM) and several ML algorithms, datasets, model validation strategies, and tuning techniques employed in previous research is carried out in the present study. Besides, the current review also summarizes DCM and ML models to predict TMC and recognize the determinants of TB in an urban area for different transport modes. The two primary goals of our study are to establish the present conceptual frameworks for the factors influencing the TMC for daily activities and to pinpoint methodological issues and limitations in previous research. With a total of 39 studies, our findings shed important light on the significance of considering factors that influence the TMC. The adjusted kernel algorithms and hyperparameter-optimized ML algorithms outperform the typical ML algorithms. RF (random forest), SVM (support vector machine), ANN (artificial neural network), and interpretable ML algorithms are the most widely used ML algorithms for the prediction of TMC where RF achieved an R2 of 0.95 and SVM achieved an accuracy of 93.18%; however, the adjusted kernel enhanced the accuracy of SVM 99.81% which shows that the interpretable algorithms outperformed the typical algorithms. The sensitivity analysis indicates that the most significant parameters influencing TMC are the age, total trip time, and the number of drivers.

Time series prediction has always been an important problem in the field of machine learning. Among them, power load forecasting plays a crucial role in identifying the behavior of photovoltaic power plants and regulating their control strategies. Traditional power load forecasting often has poor feature extraction performance for long time series. In this paper, a new deep learning framework Residual Stacked Temporal Long Short-Term Memory (RST-LSTM) is proposed, which combines wavelet decomposition and time convolutional memory network to solve the problem of feature extraction for long sequences. The network framework of RST-LSTM consists of two parts: one is a stacked time convolutional memory unit module for global and local feature extraction, and the other is a residual combination optimization module to reduce model redundancy. Finally, this paper demonstrates through various experimental indicators that RST-LSTM achieves significant performance improvements in both overall and local prediction accuracy compared to some state-of-the-art baseline methods.

The pharmaceutical industry increasingly values medicinal plants due to their perceived safety and cost-effectiveness compared to modern drugs. Throughout the extensive history of medicinal plant usage, various plant parts, including flowers, leaves, and roots, have been acknowledged for their healing properties and employed in plant identification. Leaf images, however, stand out as the preferred and easily accessible source of information. Manual plant identification by plant taxonomists is intricate, time-consuming, and prone to errors, relying heavily on human perception. Artificial intelligence (AI) techniques offer a solution by automating plant recognition processes. This study thoroughly examines cutting-edge AI approaches for leaf image-based plant identification, drawing insights from literature across renowned repositories. This paper critically summarizes relevant literature based on AI algorithms, extracted features, and results achieved. Additionally, it analyzes extensively used datasets in automated plant classification research. It also offers deep insights into implemented techniques and methods employed for medicinal plant recognition. Moreover, this rigorous review study discusses opportunities and challenges in employing these AI-based approaches. Furthermore, in-depth statistical findings and lessons learned from this survey are highlighted with novel research areas with the aim of offering insights to the readers and motivating new research directions. This review is expected to serve as a foundational resource for future researchers in the field of AI-based identification of medicinal plants.

This comprehensive review examines the structural, mechanical, electronic, and thermodynamic properties of Mg-Li-Al alloys, focusing on their corrosion resistance and mechanical performance enhancement. Utilizing first-principles calculations based on Density Functional Theory (DFT) and the quasi-harmonic approximation (QHA), the combined properties of the Mg-Li-Al phase are explored, revealing superior incompressibility, shear resistance, and stiffness compared to individual elements. The review highlights the brittleness of the alloy, supported by B/G ratios, Cauchy pressures, and Poisson’s ratios. Electronic structure analysis shows metallic behavior with varied covalent bonding characteristics, while Mulliken population analysis emphasizes significant electron transfer within the alloy. This paper also studied thermodynamic properties, including Debye temperature, heat capacity, enthalpy, free energy, and entropy, which are precisely examined, highlighting the Mg-Li-Al phase sensitive to thermal conductivity and thermal performance potential. Phonon density of states (PHDOS) confirms dynamic stability, while anisotropic sound velocities reveal elastic anisotropies. This comprehensive review not only consolidates the current understanding of the Mg-Li-Al alloy’s properties but also proposes innovative strategies for enhancing corrosion resistance. Among these strategies is the introduction of a corrosion barrier akin to the Mg-Li-Al network, which holds promise for advancing both the applications and performance of these alloys. This review serves as a crucial foundation for future research aimed at optimizing alloy design and processing methods.

Image captioning has gained increasing attention in recent years. Visual characteristics found in input images play a crucial role in generating high-quality captions. Prior studies have used visual attention mechanisms to dynamically focus on localized regions of the input image, improving the effectiveness of identifying relevant image regions at each step of caption generation. However, providing image captioning models with the capability of selecting the most relevant visual features from the input image and attending to them can significantly improve the utilization of these features. Consequently, this leads to enhanced captioning network performance. In light of this, we present an image captioning framework that efficiently exploits the extracted representations of the image. Our framework comprises three key components: the Visual Feature Detector module (VFD), the Visual Feature Visual Attention module (VFVA), and the language model. The VFD module is responsible for detecting a subset of the most pertinent features from the local visual features, creating an updated visual features matrix. Subsequently, the VFVA directs its attention to the visual features matrix generated by the VFD, resulting in an updated context vector employed by the language model to generate an informative description. Integrating the VFD and VFVA modules introduces an additional layer of processing for the visual features, thereby contributing to enhancing the image captioning model’s performance. Using the MS-COCO dataset, our experiments show that the proposed framework competes well with state-of-the-art methods, effectively leveraging visual representations to improve performance. The implementation code can be found here: https://github.com/althobhani/VFDICM (accessed on 30 July 2024).

The proposed study focuses on the critical issue of corrosion, which leads to significant economic losses and safety risks worldwide. A key area of emphasis is the accuracy of corrosion detection methods. While recent studies have made progress, a common challenge is the low accuracy of existing detection models. These models often struggle to reliably identify corrosion tendencies, which are crucial for minimizing industrial risks and optimizing resource use. The proposed study introduces an innovative approach that significantly improves the accuracy of corrosion detection using a convolutional neural network (CNN), as well as two pretrained models, namely YOLOv8 and EfficientNetB0. By leveraging advanced technologies and methodologies, we have achieved high accuracies in identifying and managing the hazards associated with corrosion across various industrial settings. This advancement not only supports the overarching goals of enhancing safety and efficiency, but also sets a new benchmark for future research in the field. The results demonstrate a significant improvement in the ability to detect and mitigate corrosion-related concerns, providing a more accurate and comprehensive solution for industries facing these challenges. Both CNN and EfficientNetB0 exhibited 100% accuracy, precision, recall, and F1-score, followed by YOLOv8 with respective metrics of 95%, 100%, 90%, and 94.74%. Our approach outperformed state-of-the-art with similar datasets and methodologies.

Cybercriminals often use fraudulent emails and fictitious email accounts to deceive individuals into disclosing confidential information, a practice known as phishing. This study utilizes three distinct methodologies, Term Frequency-Inverse Document Frequency, Word2Vec, and Bidirectional Encoder Representations from Transformers, to evaluate the effectiveness of various machine learning algorithms in detecting phishing attacks. The study uses feature extraction methods to assess the performance of Logistic Regression, Decision Tree, Random Forest, and Multilayer Perceptron algorithms. The best results for each classifier using Term Frequency-Inverse Document Frequency were Multilayer Perceptron (Precision: 0.98, Recall: 0.98, F1-score: 0.98, Accuracy: 0.98). Word2Vec’s best results were Multilayer Perceptron (Precision: 0.98, Recall: 0.98, F1-score: 0.98, Accuracy: 0.98). The highest performance was achieved using the Bidirectional Encoder Representations from the Transformers model, with Precision, Recall, F1-score, and Accuracy all reaching 0.99. This study highlights how advanced pre-trained models, such as Bidirectional Encoder Representations from Transformers, can significantly enhance the accuracy and reliability of fraud detection systems.

In numerous real-world healthcare applications, handling incomplete medical data poses significant challenges for missing value imputation and subsequent clustering or classification tasks. Traditional approaches often rely on statistical methods for imputation, which may yield suboptimal results and be computationally intensive. This paper aims to integrate imputation and clustering techniques to enhance the classification of incomplete medical data with improved accuracy. Conventional classification methods are ill-suited for incomplete medical data. To enhance efficiency without compromising accuracy, this paper introduces a novel approach that combines imputation and clustering for the classification of incomplete data. Initially, the linear interpolation imputation method alongside an iterative Fuzzy c-means clustering method is applied and followed by a classification algorithm. The effectiveness of the proposed approach is evaluated using multiple performance metrics, including accuracy, precision, specificity, and sensitivity. The encouraging results demonstrate that our proposed method surpasses classical approaches across various performance criteria.

By integrating deep neural networks with reinforcement learning, the Double Deep Q Network (DDQN) algorithm overcomes the limitations of Q-learning in handling continuous spaces and is widely applied in the path planning of mobile robots. However, the traditional DDQN algorithm suffers from sparse rewards and inefficient utilization of high-quality data. Targeting those problems, an improved DDQN algorithm based on average Q-value estimation and reward redistribution was proposed. First, to enhance the precision of the target Q-value, the average of multiple previously learned Q-values from the target Q network is used to replace the single Q-value from the current target Q network. Next, a reward redistribution mechanism is designed to overcome the sparse reward problem by adjusting the final reward of each action using the round reward from trajectory information. Additionally, a reward-prioritized experience selection method is introduced, which ranks experience samples according to reward values to ensure frequent utilization of high-quality data. Finally, simulation experiments are conducted to verify the effectiveness of the proposed algorithm in fixed-position scenario and random environments. The experimental results show that compared to the traditional DDQN algorithm, the proposed algorithm achieves shorter average running time, higher average return and fewer average steps. The performance of the proposed algorithm is improved by 11.43% in the fixed scenario and 8.33% in random environments. It not only plans economic and safe paths but also significantly improves efficiency and generalization in path planning, making it suitable for widespread application in autonomous navigation and industrial automation.

In the world of wireless sensor networks (WSNs), optimizing performance and extending network lifetime are critical goals. In this paper, we propose a new model called DTLR-Net (Deep Temporal LSTM Regression Network) that employs long-short-term memory and is effective for long-term dependencies. Mobile sinks can move in arbitrary patterns, so the model employs long short-term memory (LSTM) networks to handle such movements. The parameters were initialized iteratively, and each node updated its position, mobility level, and other important metrics at each turn, with key measurements including active or inactive node ratio, energy consumption per cycle, received packets for each node, contact time, and interconnect time between nodes, among others. These metrics aid in determining whether the model can remain stable under a variety of conditions. Furthermore, in addition to focusing on stability and security, these measurements assist us in predicting future node behaviors as well as how the network operates. The results show that the proposed model outperformed all other models by achieving a lifetime of 493.5 s for a 400-node WSN that persisted through 750 rounds, whereas other models could not reach this value and were significantly lower. This research has many implications, and one way to improve network performance dependability and sustainability is to incorporate deep learning approaches into WSN dynamics.