Pub Date : 2025-03-05DOI: 10.1016/j.engappai.2025.110351
Jinjian Lin, Kai Xie
Information fusion technology is crucial in intricate information systems, and Dempster–Shafer evidence(DSE) theory plays a significant role in it. However, most of the current research focuses on improving the conflict measurement method of high-conflict evidence in the DSE theory framework, while ignoring the comprehensive consideration of multiple conflicts of complex information. Considering the generality of graph network to complex system modeling, novel evidence measurement factors (EMF) and weighted Graph Convolution Network Dempster–Shafer evidence (wGCNDS) combination method, are proposed to optimize the combination of conflict evidence from the perspective of graph network. By constructing a weighted graph network, information transmission is realized and information fusion of associated nodes is completed. Numerical examples and real datasets verify the effectiveness and performance of wGCNDS.
{"title":"A weighted graph network-based method for combining conflicting evidence","authors":"Jinjian Lin, Kai Xie","doi":"10.1016/j.engappai.2025.110351","DOIUrl":"10.1016/j.engappai.2025.110351","url":null,"abstract":"<div><div>Information fusion technology is crucial in intricate information systems, and Dempster–Shafer evidence(DSE) theory plays a significant role in it. However, most of the current research focuses on improving the conflict measurement method of high-conflict evidence in the DSE theory framework, while ignoring the comprehensive consideration of multiple conflicts of complex information. Considering the generality of graph network to complex system modeling, novel evidence measurement factors (EMF) and weighted Graph Convolution Network Dempster–Shafer evidence (wGCNDS) combination method, are proposed to optimize the combination of conflict evidence from the perspective of graph network. By constructing a weighted graph network, information transmission is realized and information fusion of associated nodes is completed. Numerical examples and real datasets verify the effectiveness and performance of wGCNDS.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"148 ","pages":"Article 110351"},"PeriodicalIF":7.5,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549628","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-05DOI: 10.1016/j.engappai.2025.110368
Yue Wang , Jie Gao , Junchan Zhao , Xingyu Wu
For a class of higher-order nonlinear system control problems with time-varying actuator failures and external disturbances, this paper designs efficient control strategies that allow the system to be stabilized in a predefined time. First, for such systems, this paper designs an effective predefined-time control strategy using the backstepping control method combined with the adaptive radial basis neural network technique, which makes the stabilization time of the system simple and adjustable. Secondly, while using the command filtering technique to solve the “complexity explosion” problem in the design of controllers for high-order nonlinear systems, this paper designs a novel predefined-time filtering error compensation mechanism to eliminate the impact of filtering errors on the stability of the system. Finally, an event-triggered mechanism is introduced, which effectively saves the communication resources. The effectiveness of the control strategy proposed in this paper is demonstrated by the simulation experiments.
{"title":"Event-triggered predefined-time tracking control for high-order nonlinear systems with time-varying actuator failures and uncertain disturbances","authors":"Yue Wang , Jie Gao , Junchan Zhao , Xingyu Wu","doi":"10.1016/j.engappai.2025.110368","DOIUrl":"10.1016/j.engappai.2025.110368","url":null,"abstract":"<div><div>For a class of higher-order nonlinear system control problems with time-varying actuator failures and external disturbances, this paper designs efficient control strategies that allow the system to be stabilized in a predefined time. First, for such systems, this paper designs an effective predefined-time control strategy using the backstepping control method combined with the adaptive radial basis neural network technique, which makes the stabilization time of the system simple and adjustable. Secondly, while using the command filtering technique to solve the “complexity explosion” problem in the design of controllers for high-order nonlinear systems, this paper designs a novel predefined-time filtering error compensation mechanism to eliminate the impact of filtering errors on the stability of the system. Finally, an event-triggered mechanism is introduced, which effectively saves the communication resources. The effectiveness of the control strategy proposed in this paper is demonstrated by the simulation experiments.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"148 ","pages":"Article 110368"},"PeriodicalIF":7.5,"publicationDate":"2025-03-05","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549630","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-04DOI: 10.1016/j.engappai.2025.110458
Sarmed Wahab , Babatunde Abiodun Salami , Hassan Danish , Saad Nisar , Ali H. AlAteah , Ali Alsubeai
The interfacial bond strength between fiber-reinforced polymer (FRP) sheets and concrete is crucial for structural design. This study presented a novel approach using ensemble learning models to predict bond strength and analyze input parameters' influence. No previous research used gene expression programming (GEP) for developing bond strength models in single shear tests. This research introduced GEP to develop an expression for estimating bond strength, comparing its performance with existing empirical models used in design codes.
Six ensemble machine learning models were tested: extreme gradient boosting (XGBoost), light gradient boosting (LightGBM), categorical boosting (CatBoost), adaptive boosting (AdaBoost), random forest (RF), and explainable boosting machine (EBM), using 855 samples. CatBoost demonstrated superior performance with R2 = 0.98, RMSE = 1.61 kN, and MAE = 1.18 kN. The study utilized EBM's interpretability for parametric analysis through local and global explanations. Results showed FRP material and geometric properties had greater impact on bond strength than concrete properties. The novel GEP-developed empirical expression achieved higher accuracy compared to existing empirical models, with R2 = 0.812, RMSE = 4.63 kN, and MAE = 3.58 kN. The GEP model primarily relied on FRP's material and geometric properties, aligning with parametric analysis findings. Based on the results, both the CatBoost ensemble learning model and GEP model are recommended for estimating FRP-concrete interfacial bond strength.
{"title":"A hybrid machine learning approach for predicting fiber-reinforced polymer-concrete interface bond strength","authors":"Sarmed Wahab , Babatunde Abiodun Salami , Hassan Danish , Saad Nisar , Ali H. AlAteah , Ali Alsubeai","doi":"10.1016/j.engappai.2025.110458","DOIUrl":"10.1016/j.engappai.2025.110458","url":null,"abstract":"<div><div>The interfacial bond strength between fiber-reinforced polymer (FRP) sheets and concrete is crucial for structural design. This study presented a novel approach using ensemble learning models to predict bond strength and analyze input parameters' influence. No previous research used gene expression programming (GEP) for developing bond strength models in single shear tests. This research introduced GEP to develop an expression for estimating bond strength, comparing its performance with existing empirical models used in design codes.</div><div>Six ensemble machine learning models were tested: extreme gradient boosting (XGBoost), light gradient boosting (LightGBM), categorical boosting (CatBoost), adaptive boosting (AdaBoost), random forest (RF), and explainable boosting machine (EBM), using 855 samples. CatBoost demonstrated superior performance with R2 = 0.98, RMSE = 1.61 kN, and MAE = 1.18 kN. The study utilized EBM's interpretability for parametric analysis through local and global explanations. Results showed FRP material and geometric properties had greater impact on bond strength than concrete properties. The novel GEP-developed empirical expression achieved higher accuracy compared to existing empirical models, with R2 = 0.812, RMSE = 4.63 kN, and MAE = 3.58 kN. The GEP model primarily relied on FRP's material and geometric properties, aligning with parametric analysis findings. Based on the results, both the CatBoost ensemble learning model and GEP model are recommended for estimating FRP-concrete interfacial bond strength.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"148 ","pages":"Article 110458"},"PeriodicalIF":7.5,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549694","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-04DOI: 10.1016/j.engappai.2025.110456
Hanxiang Wang , Yanfen Li , Tan N. Nguyen , L. Minh Dang
Traditional network architectures struggled with a uniform approach to receptive field (RF) sizes, leading to suboptimal performance across scales. Although recent advances have addressed the problem by utilizing different RF sizes, a balance between accuracy and complexity remains elusive. In addition, the existing group attention mechanism that simply uses the squeeze-and-excitation method neglects the spatial position information in the feature selection and fusion process. Therefore, this research introduces a lightweight and efficient architecture named Split-Dense Adaptive Network (SDANet) to cope with these limitations. In the proposed network, a residual-like multi-kernel method is implemented to enable better feature extraction under diverse RF sizes. Next, a new grouped attention module processes features dynamically and highlight the location information. Also, the constructed feature augmentation structure strengthens the model's representation. Furthermore, a new channel split and merge strategy is utilized for computation reduction. Compared with state-of-the-art methods, our model achieved better generalization ability, less computational complexity, and superior precision based on various public datasets. The introduced network shows a promising general applicability in the field of computer vision, and further inspires research on supervised deep learning.
{"title":"Residual-like multi-kernel block and dynamic attention for deep neural networks","authors":"Hanxiang Wang , Yanfen Li , Tan N. Nguyen , L. Minh Dang","doi":"10.1016/j.engappai.2025.110456","DOIUrl":"10.1016/j.engappai.2025.110456","url":null,"abstract":"<div><div>Traditional network architectures struggled with a uniform approach to receptive field (RF) sizes, leading to suboptimal performance across scales. Although recent advances have addressed the problem by utilizing different RF sizes, a balance between accuracy and complexity remains elusive. In addition, the existing group attention mechanism that simply uses the squeeze-and-excitation method neglects the spatial position information in the feature selection and fusion process. Therefore, this research introduces a lightweight and efficient architecture named Split-Dense Adaptive Network (SDANet) to cope with these limitations. In the proposed network, a residual-like multi-kernel method is implemented to enable better feature extraction under diverse RF sizes. Next, a new grouped attention module processes features dynamically and highlight the location information. Also, the constructed feature augmentation structure strengthens the model's representation. Furthermore, a new channel split and merge strategy is utilized for computation reduction. Compared with state-of-the-art methods, our model achieved better generalization ability, less computational complexity, and superior precision based on various public datasets. The introduced network shows a promising general applicability in the field of computer vision, and further inspires research on supervised deep learning.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"148 ","pages":"Article 110456"},"PeriodicalIF":7.5,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143535296","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-04DOI: 10.1016/j.engappai.2025.110393
Zhaokang Sheng , Tingqiang Song , Jiale Song , Yalin Liu , Peng Ren
Path planning is central to the operation of intelligent systems such as robots, drones, and autonomous vehicles, where path performance and time efficiency directly impact overall system performance. Although sampling-based path planning methods have achieved significant success in this field, their performance remains limited in crowded environments. This paper combines and improves the bidirectional exploration method of BI-RRT* (Bidirectional Rapidly-exploring Random Tree Star) and the expansion guidance of APF-RRT* (Artificial Potential Field Rapidly-exploring Random Tree Star), proposing a bidirectional rapidly exploring random tree algorithm based on adaptive mechanisms and artificial potential fields (AB-APF-RRT*). This method improves both the sampling and expansion methods of RRT*(Rapidly-exploring Random Tree Star) . In terms of sampling, the probabilities in different regions are modified using the line connecting the start and goal points, and dynamic goal bias and opposing bias strategies are introduced to guide the trees towards the target and each other. In terms of expansion, based on the bidirectional exploration of the two trees, optimized artificial potential fields and ray-casting navigation strategies are applied to guide the trees towards the goal while avoiding obstacles and dynamically adjusting the step size. To enhance the smoothness of the path, a cubic spline interpolation method is further applied. Ultimately, a comparison with several popular sampling-based path planning algorithms demonstrates that this method excels in both performance and time efficiency.
{"title":"Bidirectional rapidly exploring random tree path planning algorithm based on adaptive strategies and artificial potential fields","authors":"Zhaokang Sheng , Tingqiang Song , Jiale Song , Yalin Liu , Peng Ren","doi":"10.1016/j.engappai.2025.110393","DOIUrl":"10.1016/j.engappai.2025.110393","url":null,"abstract":"<div><div>Path planning is central to the operation of intelligent systems such as robots, drones, and autonomous vehicles, where path performance and time efficiency directly impact overall system performance. Although sampling-based path planning methods have achieved significant success in this field, their performance remains limited in crowded environments. This paper combines and improves the bidirectional exploration method of BI-RRT* (Bidirectional Rapidly-exploring Random Tree Star) and the expansion guidance of APF-RRT* (Artificial Potential Field Rapidly-exploring Random Tree Star), proposing a bidirectional rapidly exploring random tree algorithm based on adaptive mechanisms and artificial potential fields (AB-APF-RRT*). This method improves both the sampling and expansion methods of RRT*(Rapidly-exploring Random Tree Star) . In terms of sampling, the probabilities in different regions are modified using the line connecting the start and goal points, and dynamic goal bias and opposing bias strategies are introduced to guide the trees towards the target and each other. In terms of expansion, based on the bidirectional exploration of the two trees, optimized artificial potential fields and ray-casting navigation strategies are applied to guide the trees towards the goal while avoiding obstacles and dynamically adjusting the step size. To enhance the smoothness of the path, a cubic spline interpolation method is further applied. Ultimately, a comparison with several popular sampling-based path planning algorithms demonstrates that this method excels in both performance and time efficiency.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"148 ","pages":"Article 110393"},"PeriodicalIF":7.5,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549696","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-04DOI: 10.1016/j.engappai.2025.110371
Maedeh H. Toosi, Mahdi Mohammadi-nasab, Siamak Mohammadi , Mostafa E. Salehi
Atrial Fibrillation (AF) detection from electrocardiogram (ECG) signals has become vital for early diagnosis and continuous management in personal healthcare. Developing machine learning models for AF classification is challenging due to the various sources of ECG signals across datasets, which affects model generalization. To address this issue, we have proposed a lightweight transformer-based model using two public datasets and have evaluated its cross-domain generalization. The model has been further tested on a distinct dataset from the China Physiological Signal Challenge (CPSC) 2018. As an additional validation step, we have also tested it on real-world ECG data collected from wearable home monitoring devices, emphasizing its adaptability to diverse recording conditions. Despite these challenges, our transformer model achieves an accuracy of 85.6% in wearable data, significantly improving AF detection by analyzing complex ECG patterns and helping early diagnosis and continuous management of AF. Additionally, we have implemented a quantization technique that compresses the model to low bit precision for deployment on resource-constrained devices. The model, quantized to 3 bits for both weights and activations, maintains its accuracy while achieving a model size that is more than 9x smaller. Moreover, a fully binarized version with 5-bit quantized activations offers a 19x reduction in size, with only a minimal accuracy decrease of less than 2%.
{"title":"Efficient quantized transformer for atrial fibrillation detection in cross-domain datasets","authors":"Maedeh H. Toosi, Mahdi Mohammadi-nasab, Siamak Mohammadi , Mostafa E. Salehi","doi":"10.1016/j.engappai.2025.110371","DOIUrl":"10.1016/j.engappai.2025.110371","url":null,"abstract":"<div><div>Atrial Fibrillation (AF) detection from electrocardiogram (ECG) signals has become vital for early diagnosis and continuous management in personal healthcare. Developing machine learning models for AF classification is challenging due to the various sources of ECG signals across datasets, which affects model generalization. To address this issue, we have proposed a lightweight transformer-based model using two public datasets and have evaluated its cross-domain generalization. The model has been further tested on a distinct dataset from the China Physiological Signal Challenge (CPSC) 2018. As an additional validation step, we have also tested it on real-world ECG data collected from wearable home monitoring devices, emphasizing its adaptability to diverse recording conditions. Despite these challenges, our transformer model achieves an accuracy of 85.6% in wearable data, significantly improving AF detection by analyzing complex ECG patterns and helping early diagnosis and continuous management of AF. Additionally, we have implemented a quantization technique that compresses the model to low bit precision for deployment on resource-constrained devices. The model, quantized to 3 bits for both weights and activations, maintains its accuracy while achieving a model size that is more than 9x smaller. Moreover, a fully binarized version with 5-bit quantized activations offers a 19x reduction in size, with only a minimal accuracy decrease of less than 2%.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"148 ","pages":"Article 110371"},"PeriodicalIF":7.5,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549693","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-04DOI: 10.1016/j.engappai.2025.110448
Tao Liu , Yibin Liu , Jian Chen , Jin Gong
During hydro-zinc smelting, the cathode plates are attached by with residual zinc or discarded due to damaged insulation strips and edging strips. Such defects limit the recycling of cathode plates. Current manual observation leads to low accuracy and speed of recognition owing to perception biases. Therefore, this work applied computer vision and deep learning semantic segmentation technology to realize the defect recognition of cathode plates. Firstly, a semantic segmentation dataset on cathode plates was constructed for training and testing the model. Then a network of attention mechanism and multiscale feature fusion (AMNet) was proposed to detect the defects. In AMNet, the encoder-decoder jump connection architecture was designed to fuse low-level and high-level features. A channel attention module was incorporated to enhance focus on the channels with important information, and the newly proposed multiscale feature extraction module was used to solve the problem of target multiscale capture. Through related parameter selection experiments, the final AMNet achieved 95.12% and 97.73% for Mean Intersection over Union (MIoU) and mean pixel accuracy (MPA), respectively. These values are 3.24 and 1.74 percentage points higher than DeepLabv3+.
{"title":"Identification of zinc stripping defects from cathode plate based on deep learning","authors":"Tao Liu , Yibin Liu , Jian Chen , Jin Gong","doi":"10.1016/j.engappai.2025.110448","DOIUrl":"10.1016/j.engappai.2025.110448","url":null,"abstract":"<div><div>During hydro-zinc smelting, the cathode plates are attached by with residual zinc or discarded due to damaged insulation strips and edging strips. Such defects limit the recycling of cathode plates. Current manual observation leads to low accuracy and speed of recognition owing to perception biases. Therefore, this work applied computer vision and deep learning semantic segmentation technology to realize the defect recognition of cathode plates. Firstly, a semantic segmentation dataset on cathode plates was constructed for training and testing the model. Then a network of attention mechanism and multiscale feature fusion (AMNet) was proposed to detect the defects. In AMNet, the encoder-decoder jump connection architecture was designed to fuse low-level and high-level features. A channel attention module was incorporated to enhance focus on the channels with important information, and the newly proposed multiscale feature extraction module was used to solve the problem of target multiscale capture. Through related parameter selection experiments, the final AMNet achieved 95.12% and 97.73% for Mean Intersection over Union (MIoU) and mean pixel accuracy (MPA), respectively. These values are 3.24 and 1.74 percentage points higher than DeepLabv3+.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"148 ","pages":"Article 110448"},"PeriodicalIF":7.5,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143535298","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-04DOI: 10.1016/j.engappai.2025.110399
Priyanka D., Mala T.
Gait recognition has emerged as a crucial biometric identifier due to its non-invasive and unobtrusive characteristics. Unlike silhouette-based methods, which include appearance information, skeleton-based gait recognition offers gait data without visual clues. However, traditional models in this field often rely on handcrafted features and adjacency matrices formed from physically connected edges, posing a significant challenge in extracting semantically meaningful joints and edges. To address this challenge, a novel Semantic relationship and hierarchical Fusion-based Graph Network (SFG-Net) utilizing a Hierarchical-joint Connectivity Graph (HC-Graph) is proposed. SFG-Net divides each joint node into multiple subsets, facilitating the extraction of both proximal and distant edges, and constructs an HC-Graph to represent these edges within the semantic spaces of the human skeleton. Furthermore, a Hierarchical Attention (HA) mechanism is introduced to emphasize dominant hierarchical edge sets within the HC-Graph. The temporal dynamics of the gait data are captured using Multi-scale Temporal Convolution (MSTC). To further enhance discriminative power, features at different levels are concatenated, capturing both dynamic and structurally semantic features. Experimental results on benchmark gait recognition datasets show that the proposed SFG-Net significantly outperforms current state-of-the-art methods, exhibiting superior robustness and accuracy across various challenging scenarios.
{"title":"SFG-Net: Semantic relationship and hierarchical Fusion-based Graph Network for enhanced skeleton-based gait recognition","authors":"Priyanka D., Mala T.","doi":"10.1016/j.engappai.2025.110399","DOIUrl":"10.1016/j.engappai.2025.110399","url":null,"abstract":"<div><div>Gait recognition has emerged as a crucial biometric identifier due to its non-invasive and unobtrusive characteristics. Unlike silhouette-based methods, which include appearance information, skeleton-based gait recognition offers gait data without visual clues. However, traditional models in this field often rely on handcrafted features and adjacency matrices formed from physically connected edges, posing a significant challenge in extracting semantically meaningful joints and edges. To address this challenge, a novel Semantic relationship and hierarchical Fusion-based Graph Network (SFG-Net) utilizing a Hierarchical-joint Connectivity Graph (HC-Graph) is proposed. SFG-Net divides each joint node into multiple subsets, facilitating the extraction of both proximal and distant edges, and constructs an HC-Graph to represent these edges within the semantic spaces of the human skeleton. Furthermore, a Hierarchical Attention (HA) mechanism is introduced to emphasize dominant hierarchical edge sets within the HC-Graph. The temporal dynamics of the gait data are captured using Multi-scale Temporal Convolution (MSTC). To further enhance discriminative power, features at different levels are concatenated, capturing both dynamic and structurally semantic features. Experimental results on benchmark gait recognition datasets show that the proposed SFG-Net significantly outperforms current state-of-the-art methods, exhibiting superior robustness and accuracy across various challenging scenarios.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"148 ","pages":"Article 110399"},"PeriodicalIF":7.5,"publicationDate":"2025-03-04","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143549695","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-03DOI: 10.1016/j.engappai.2025.110344
Yuhui Zhou, Xiaojie Wu, Yiming Li, Huimin Sun, Di Fan
The velocity measurement of trunk canal and river plays an important role in agriculture and forestry irrigation scheduling, water resources management and flood prediction. Particle flow measurement technology can realize non-contact and high-precision flow measurement, but in practical application, the particle size is small, the shape is different and the dynamic change brings great challenges to the application of this method. To solve these problems, this paper proposed the surface velocity measurement method of trunk canal based on improved YOLOv8(You Only Look Once Version 8) and DeepSORT(Deep Simple Online and Realtime Tracking), and introduced tiny detection layer and channel attention mechanism to improve YOLOv8's detection capability of small targets. In DeepSORT, IBN-Net(Intent-Based Networking-Network) network structure and GIoU(Generalized Intersection over Union) matching are introduced to solve the problem of discontinuity or loss of target tracking in complex cases, which improves the accuracy and robustness of target tracking. The experimental results show that the improved YOLOv8 improves AP(Average Precision) and mAP(mean Average Precision) by nearly 5% and 0.2% respectively. The performance of the improved DeepSORT has been improved across the board, especially IDP and MOTA, which have improved by 25.2% and 5.6% respectively. The algorithm also has good accuracy in actual velocity measurement.
{"title":"Algorithm for surface flow velocity measurement in trunk canal based on improved YOLOv8 and DeepSORT","authors":"Yuhui Zhou, Xiaojie Wu, Yiming Li, Huimin Sun, Di Fan","doi":"10.1016/j.engappai.2025.110344","DOIUrl":"10.1016/j.engappai.2025.110344","url":null,"abstract":"<div><div>The velocity measurement of trunk canal and river plays an important role in agriculture and forestry irrigation scheduling, water resources management and flood prediction. Particle flow measurement technology can realize non-contact and high-precision flow measurement, but in practical application, the particle size is small, the shape is different and the dynamic change brings great challenges to the application of this method. To solve these problems, this paper proposed the surface velocity measurement method of trunk canal based on improved YOLOv8(You Only Look Once Version 8) and DeepSORT(Deep Simple Online and Realtime Tracking), and introduced tiny detection layer and channel attention mechanism to improve YOLOv8's detection capability of small targets. In DeepSORT, IBN-Net(Intent-Based Networking-Network) network structure and GIoU(Generalized Intersection over Union) matching are introduced to solve the problem of discontinuity or loss of target tracking in complex cases, which improves the accuracy and robustness of target tracking. The experimental results show that the improved YOLOv8 improves AP(Average Precision) and mAP(mean Average Precision) by nearly 5% and 0.2% respectively. The performance of the improved DeepSORT has been improved across the board, especially IDP and MOTA, which have improved by 25.2% and 5.6% respectively. The algorithm also has good accuracy in actual velocity measurement.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"148 ","pages":"Article 110344"},"PeriodicalIF":7.5,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143528958","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2025-03-03DOI: 10.1016/j.engappai.2025.110294
Ruilian Li , Deliang Zeng , Tingting Li , Yan Xie , Yong Hu , Guangming Zhang
With the fluctuation of load in coal-fired power plants, the sulfur dioxide (SO2) concentration in the flue gas changes more and more frequently, due to the large delay and inertia characteristics of wet flue gas desulfurization (WFGD) systems, the SO2 concentration in the flue gas emitted from the outlet is unstable. To accurately control the SO2 emission concentration of the desulfurization system, a single-tower double-circulation wet flue gas desulfurization (SD-WFGD) system outlet SO2 concentration prediction model was established. Firstly, considering the application of the established model in control systems and system optimization operation schemes, it is necessary to enhance the interpretability of the model. Therefore, based on the chemical reaction process of SO2 absorption in the desulfurization system, SO2 mechanism prediction models for the outlet of the absorption tower and absorber feed tank (AFT) tower were established, and parameter identification was carried out using quantum particle swarm optimization algorithm (QPSO) and historical operation data of the power station. Secondly, to improve the prediction accuracy of the mechanism model, a Convolutional Neural Network (CNN)-Long Short Term Memory (LSTM)-Attention data compensation model was established. In this process, the difficulty of manually adjusting the hyperparameters of the deep learning model was considered, during the model training process, the rime optimization algorithm was used to optimize the model hyperparameters in real time. To enable the data compensation model to obtain more data feature information, the input data was decomposed using the Variational Mode Decomposition (VMD) method and the same frequency modes were combined and reconstructed. After model training, different modal model outputs were superimposed to obtain the final compensation data. Finally, the mechanism model and data compensation model were combined to obtain a hybrid prediction model for SO2 concentration. The model validation results showed that the error indicators root mean squared error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R2) of the model are 0.2739 mg/m3, 0.8267%, and 0.9999, respectively.
{"title":"Dynamic prediction of sulfur dioxide concentration in a single-tower double-circulation desulfurization system based on chemical mechanism and deep learning","authors":"Ruilian Li , Deliang Zeng , Tingting Li , Yan Xie , Yong Hu , Guangming Zhang","doi":"10.1016/j.engappai.2025.110294","DOIUrl":"10.1016/j.engappai.2025.110294","url":null,"abstract":"<div><div>With the fluctuation of load in coal-fired power plants, the sulfur dioxide (SO<sub>2</sub>) concentration in the flue gas changes more and more frequently, due to the large delay and inertia characteristics of wet flue gas desulfurization (WFGD) systems, the SO<sub>2</sub> concentration in the flue gas emitted from the outlet is unstable. To accurately control the SO<sub>2</sub> emission concentration of the desulfurization system, a single-tower double-circulation wet flue gas desulfurization (SD-WFGD) system outlet SO<sub>2</sub> concentration prediction model was established. Firstly, considering the application of the established model in control systems and system optimization operation schemes, it is necessary to enhance the interpretability of the model. Therefore, based on the chemical reaction process of SO<sub>2</sub> absorption in the desulfurization system, SO<sub>2</sub> mechanism prediction models for the outlet of the absorption tower and absorber feed tank (AFT) tower were established, and parameter identification was carried out using quantum particle swarm optimization algorithm (QPSO) and historical operation data of the power station. Secondly, to improve the prediction accuracy of the mechanism model, a Convolutional Neural Network (CNN)-Long Short Term Memory (LSTM)-Attention data compensation model was established. In this process, the difficulty of manually adjusting the hyperparameters of the deep learning model was considered, during the model training process, the rime optimization algorithm was used to optimize the model hyperparameters in real time. To enable the data compensation model to obtain more data feature information, the input data was decomposed using the Variational Mode Decomposition (VMD) method and the same frequency modes were combined and reconstructed. After model training, different modal model outputs were superimposed to obtain the final compensation data. Finally, the mechanism model and data compensation model were combined to obtain a hybrid prediction model for SO<sub>2</sub> concentration. The model validation results showed that the error indicators root mean squared error (RMSE), mean absolute percentage error (MAPE), and coefficient of determination (R<sup>2</sup>) of the model are 0.2739 mg/m<sup>3</sup>, 0.8267%, and 0.9999, respectively.</div></div>","PeriodicalId":50523,"journal":{"name":"Engineering Applications of Artificial Intelligence","volume":"148 ","pages":"Article 110294"},"PeriodicalIF":7.5,"publicationDate":"2025-03-03","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"143528960","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":2,"RegionCategory":"计算机科学","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
扫码关注我们
求助内容：
应助结果提醒方式：
京公网安备 11010802042870号