Pub Date : 2024-01-26DOI: 10.1109/ICPECA60615.2024.10471165
Haiyang Liu, Hongliu Yang, Weihao Gao, Bo Zhang, Zichen Gao
Safety management and control in live electricity operation sites constitute a crucial assurance component for electrical safety production. As the demand for live electricity operations continues to rise, accompanied by increased complexity and difficulty, the shift from manual video analysis to intelligent control methods in on-site safety management has become imperative. In response to this, a human body posture recognition technology is proposed, utilizing YOLOv8 to establish a multi-person posture recognition model. This, combined with traditional image recognition techniques, achieves comprehensive perception of personnel states, enabling real-time management and early warning of hazards and non-standard behaviors during operations. This approach alleviates the pressure on inspection personnel and enhances the intelligence of violation recognition in live electricity operation sites.
{"title":"Research on Deep Learning-Based Recognition Technology for Violations in Live Electricity Operations","authors":"Haiyang Liu, Hongliu Yang, Weihao Gao, Bo Zhang, Zichen Gao","doi":"10.1109/ICPECA60615.2024.10471165","DOIUrl":"https://doi.org/10.1109/ICPECA60615.2024.10471165","url":null,"abstract":"Safety management and control in live electricity operation sites constitute a crucial assurance component for electrical safety production. As the demand for live electricity operations continues to rise, accompanied by increased complexity and difficulty, the shift from manual video analysis to intelligent control methods in on-site safety management has become imperative. In response to this, a human body posture recognition technology is proposed, utilizing YOLOv8 to establish a multi-person posture recognition model. This, combined with traditional image recognition techniques, achieves comprehensive perception of personnel states, enabling real-time management and early warning of hazards and non-standard behaviors during operations. This approach alleviates the pressure on inspection personnel and enhances the intelligence of violation recognition in live electricity operation sites.","PeriodicalId":518671,"journal":{"name":"2024 IEEE 4th International Conference on Power, Electronics and Computer Applications (ICPECA)","volume":"120 3","pages":"356-359"},"PeriodicalIF":0.0,"publicationDate":"2024-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530361","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-26DOI: 10.1109/ICPECA60615.2024.10471097
Yinlong Wang, Wang Dan, Juntao Ma
In order to improve the maximum warning range of radar networks, an optimization algorithm for radar network layout based on improved genetic algorithm is proposed. First, a joint early warning probability calculation model for multiple radars was established. Subsequently, a detailed process for calculating early warning using the Monte Carlo method for radar networks was presented, and an improved genetic algorithm was proposed for solving the problem. The improved genetic algorithm mainly improves on the roulette algorithm, DNA selection, crossover, and mutation. Simulation experiments show that the improved algorithm improves the convergence speed.
{"title":"A Radar Net Layout Method Based on Improved Genetic Algorithm","authors":"Yinlong Wang, Wang Dan, Juntao Ma","doi":"10.1109/ICPECA60615.2024.10471097","DOIUrl":"https://doi.org/10.1109/ICPECA60615.2024.10471097","url":null,"abstract":"In order to improve the maximum warning range of radar networks, an optimization algorithm for radar network layout based on improved genetic algorithm is proposed. First, a joint early warning probability calculation model for multiple radars was established. Subsequently, a detailed process for calculating early warning using the Monte Carlo method for radar networks was presented, and an improved genetic algorithm was proposed for solving the problem. The improved genetic algorithm mainly improves on the roulette algorithm, DNA selection, crossover, and mutation. Simulation experiments show that the improved algorithm improves the convergence speed.","PeriodicalId":518671,"journal":{"name":"2024 IEEE 4th International Conference on Power, Electronics and Computer Applications (ICPECA)","volume":"79 4","pages":"18-22"},"PeriodicalIF":0.0,"publicationDate":"2024-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530291","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-26DOI: 10.1109/ICPECA60615.2024.10471026
Dan Zhang, Guolv Zhu, Shibo Lu, Chang Li
Lane detection is one of the important tasks of the environmental patrol work in power plant. In order to improve the detection accuracy of lane, this paper proposes a tensor fusion structure RCFPN, and takes the lane detection model RESA as baseline. After the backbone feature extraction network of RESA model, RCFPN is added to construct the improved network. The experimental results prove that RCFPN has an effect on improving RESA's precision. RCFPN can not only improve the precision of RESA model, but also can be flexibly integrated into other lane detection models and other target detection models. The average detection accuracy of CULANE was increased from 75.31% to 77.76%. The F1 score, accurary, FP, FN are better than the original model in the Tusimple data set.
{"title":"Lane Detection Based on Improved RESA in Power Plant","authors":"Dan Zhang, Guolv Zhu, Shibo Lu, Chang Li","doi":"10.1109/ICPECA60615.2024.10471026","DOIUrl":"https://doi.org/10.1109/ICPECA60615.2024.10471026","url":null,"abstract":"Lane detection is one of the important tasks of the environmental patrol work in power plant. In order to improve the detection accuracy of lane, this paper proposes a tensor fusion structure RCFPN, and takes the lane detection model RESA as baseline. After the backbone feature extraction network of RESA model, RCFPN is added to construct the improved network. The experimental results prove that RCFPN has an effect on improving RESA's precision. RCFPN can not only improve the precision of RESA model, but also can be flexibly integrated into other lane detection models and other target detection models. The average detection accuracy of CULANE was increased from 75.31% to 77.76%. The F1 score, accurary, FP, FN are better than the original model in the Tusimple data set.","PeriodicalId":518671,"journal":{"name":"2024 IEEE 4th International Conference on Power, Electronics and Computer Applications (ICPECA)","volume":"4 3-4","pages":"108-112"},"PeriodicalIF":0.0,"publicationDate":"2024-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530131","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-26DOI: 10.1109/ICPECA60615.2024.10471151
Zihao Yan, Huishan Zhang, Liping Li
Aiming at the problems of the dense target distribution, poor positioning ability to pick robots, and inaccurate ripeness judgment in the picking orchard scene, this paper proposes an apple image recognition model with a high recognition rate, high speed, and high accuracy, which can effectively analyze the data of quantity, location, ripeness and quality estimation in the apple image. Firstly, the Faster R-CNN network is improved by introducing Efficient Channel Attention (ECA) and multi-scale fusion feature pyramid (FPN) for fruit detection and recognition localization. Then the distance transform-based watershed algorithm is used for image segmentation to fit the apple edge image while combining with the fitted circle determination algorithm to establish a mathematical model for apple volume estimation to calculate the quantity as well as the quality of apples. Finally, the apples are classified into four categories according to their ripeness, and the improved Faster R-CNN network is used to improve the ripeness detection effect, and the results show that the average fruit recognition accuracy of the improved method proposed in this paper is 95.42%, which significantly improves the accuracy of fruit detection.
{"title":"An Integrated Target Recognition Method Based on Improved Faster-RCNN for Apple Detection, Counting, Localization, and Quality Estimation","authors":"Zihao Yan, Huishan Zhang, Liping Li","doi":"10.1109/ICPECA60615.2024.10471151","DOIUrl":"https://doi.org/10.1109/ICPECA60615.2024.10471151","url":null,"abstract":"Aiming at the problems of the dense target distribution, poor positioning ability to pick robots, and inaccurate ripeness judgment in the picking orchard scene, this paper proposes an apple image recognition model with a high recognition rate, high speed, and high accuracy, which can effectively analyze the data of quantity, location, ripeness and quality estimation in the apple image. Firstly, the Faster R-CNN network is improved by introducing Efficient Channel Attention (ECA) and multi-scale fusion feature pyramid (FPN) for fruit detection and recognition localization. Then the distance transform-based watershed algorithm is used for image segmentation to fit the apple edge image while combining with the fitted circle determination algorithm to establish a mathematical model for apple volume estimation to calculate the quantity as well as the quality of apples. Finally, the apples are classified into four categories according to their ripeness, and the improved Faster R-CNN network is used to improve the ripeness detection effect, and the results show that the average fruit recognition accuracy of the improved method proposed in this paper is 95.42%, which significantly improves the accuracy of fruit detection.","PeriodicalId":518671,"journal":{"name":"2024 IEEE 4th International Conference on Power, Electronics and Computer Applications (ICPECA)","volume":"31 3-4","pages":"726-731"},"PeriodicalIF":0.0,"publicationDate":"2024-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530107","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-26DOI: 10.1109/ICPECA60615.2024.10471135
Xiaosong Zou
This paper introduces a real-time face detection technology based on TMS320C6201. Through the confidential communication between each subsystem, the synchronization of each subsystem is completed, and the real-time face recognition, feature code extraction, and the closest face matching are carried out. Firstly, Grabcut foreground extraction method is used for pre-background segmentation of recognized images to reduce external interference, and then face detection and identification are carried out according to the segmentation effect. A parallel MPI program is developed by transforming the traditional serialization-based face information updating method into a parallel method. This paper applies existing MPI-based methods and existing web-based facial information acquisition methods to improve the efficiency of existing face recognition technologies. It realizes the distributed processing of the update algorithm in the original face recognition system and enhances the ability of the system to process a large amount of data to achieve the purpose of improving the system performance. The experimental results show that the system combining grab cut and Adaboost algorithm can improve the recognition rate and detection rate, and the recognition speed is faster.
{"title":"Research on Face Recognition System Based on Intelligent Machine Learning Algorithm","authors":"Xiaosong Zou","doi":"10.1109/ICPECA60615.2024.10471135","DOIUrl":"https://doi.org/10.1109/ICPECA60615.2024.10471135","url":null,"abstract":"This paper introduces a real-time face detection technology based on TMS320C6201. Through the confidential communication between each subsystem, the synchronization of each subsystem is completed, and the real-time face recognition, feature code extraction, and the closest face matching are carried out. Firstly, Grabcut foreground extraction method is used for pre-background segmentation of recognized images to reduce external interference, and then face detection and identification are carried out according to the segmentation effect. A parallel MPI program is developed by transforming the traditional serialization-based face information updating method into a parallel method. This paper applies existing MPI-based methods and existing web-based facial information acquisition methods to improve the efficiency of existing face recognition technologies. It realizes the distributed processing of the update algorithm in the original face recognition system and enhances the ability of the system to process a large amount of data to achieve the purpose of improving the system performance. The experimental results show that the system combining grab cut and Adaboost algorithm can improve the recognition rate and detection rate, and the recognition speed is faster.","PeriodicalId":518671,"journal":{"name":"2024 IEEE 4th International Conference on Power, Electronics and Computer Applications (ICPECA)","volume":"43 3","pages":"1028-1032"},"PeriodicalIF":0.0,"publicationDate":"2024-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530502","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-26DOI: 10.1109/ICPECA60615.2024.10471053
Liu Xia, Yuanqing Wang, Xueling Li
This paper introduces a multi-view sub-pixel arrangement method based on multi-view autostereoscopic display. The traditional autostereoscopic display using lenticular sheet covers an integer number of sub-pixels and arranges the sub-pixels directly according to the order of view. This method is a generalized algorithm for arranging multi-view sub-pixels under conditions for slanted gratings. According to the number of views, the number of grating in the display unit and other related parameters, the optimal arrangement is calculated. Finally, experiments with 32 views and 3 sets of grating in the display unit verify that the sub-pixel arrangement results obtained by the algorithm can achieve the correct exit pupil distribution and a continuous parallax effect.
{"title":"Sub-Pixel Arrangement Algorithm Based on Multi-View Autostereoscopic Three-Dimension Display Using a Slanted Lenticular Sheet","authors":"Liu Xia, Yuanqing Wang, Xueling Li","doi":"10.1109/ICPECA60615.2024.10471053","DOIUrl":"https://doi.org/10.1109/ICPECA60615.2024.10471053","url":null,"abstract":"This paper introduces a multi-view sub-pixel arrangement method based on multi-view autostereoscopic display. The traditional autostereoscopic display using lenticular sheet covers an integer number of sub-pixels and arranges the sub-pixels directly according to the order of view. This method is a generalized algorithm for arranging multi-view sub-pixels under conditions for slanted gratings. According to the number of views, the number of grating in the display unit and other related parameters, the optimal arrangement is calculated. Finally, experiments with 32 views and 3 sets of grating in the display unit verify that the sub-pixel arrangement results obtained by the algorithm can achieve the correct exit pupil distribution and a continuous parallax effect.","PeriodicalId":518671,"journal":{"name":"2024 IEEE 4th International Conference on Power, Electronics and Computer Applications (ICPECA)","volume":"48 3","pages":"152-156"},"PeriodicalIF":0.0,"publicationDate":"2024-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530304","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-26DOI: 10.1109/ICPECA60615.2024.10471008
Wanfeng Ji, Cheng Li, Yaoqing Zhang
If you check the Index mirror of marine sextant from different angle, you will get different result. To solve the problem of marine sextant, in this article the checking method and principle of index mirror is analyzed, the problem existing in the design of marine sextant is studied. The method of simulation is adopted to calculate the vertical error of index mirror, and calculate the corresponding height measurement error, the poor distribution is formulated. The result to improve the measurement precision has very important practical significance.
{"title":"Performance Analysis and Simulation Calculation of Marine Sextant","authors":"Wanfeng Ji, Cheng Li, Yaoqing Zhang","doi":"10.1109/ICPECA60615.2024.10471008","DOIUrl":"https://doi.org/10.1109/ICPECA60615.2024.10471008","url":null,"abstract":"If you check the Index mirror of marine sextant from different angle, you will get different result. To solve the problem of marine sextant, in this article the checking method and principle of index mirror is analyzed, the problem existing in the design of marine sextant is studied. The method of simulation is adopted to calculate the vertical error of index mirror, and calculate the corresponding height measurement error, the poor distribution is formulated. The result to improve the measurement precision has very important practical significance.","PeriodicalId":518671,"journal":{"name":"2024 IEEE 4th International Conference on Power, Electronics and Computer Applications (ICPECA)","volume":"118 1","pages":"294-298"},"PeriodicalIF":0.0,"publicationDate":"2024-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530465","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-26DOI: 10.1109/ICPECA60615.2024.10471073
Meng Yibo, Hong Dehua, Zhang Min, Li Ming, Chai Wujun, Li Bin
Operation and maintenance system is an important platform to support the deployment, operation and evaluation of distributed simulation system. The operation monitoring tool is to manage the simulation node, control the operation process, feedback the operation results, and evaluate the operation effect. Then this paper mainly describes the composition of the operation management system. Through the research of distributed simulation test, a distributed simulation test management platform based on virtual instrument is proposed. The corresponding design ideas are given based on its functional requirements. The architecture, module composition and overall working process of the system are described. The working mechanism, design and implementation of node management, auxiliary analysis, instruction management and condition monitoring are analyzed. Multi-dimensional multi-source heterogeneous fault data are used as training samples to achieve automatic detection of equipment operating conditions. Reinforcement learning theory is applied to equipment fault diagnosis to achieve automatic judgment of test data, so as to establish the active early warning mechanism of equipment.
{"title":"Design of Distributed Simulation Operation and Maintenance Management System Based on Artificial Intelligence","authors":"Meng Yibo, Hong Dehua, Zhang Min, Li Ming, Chai Wujun, Li Bin","doi":"10.1109/ICPECA60615.2024.10471073","DOIUrl":"https://doi.org/10.1109/ICPECA60615.2024.10471073","url":null,"abstract":"Operation and maintenance system is an important platform to support the deployment, operation and evaluation of distributed simulation system. The operation monitoring tool is to manage the simulation node, control the operation process, feedback the operation results, and evaluate the operation effect. Then this paper mainly describes the composition of the operation management system. Through the research of distributed simulation test, a distributed simulation test management platform based on virtual instrument is proposed. The corresponding design ideas are given based on its functional requirements. The architecture, module composition and overall working process of the system are described. The working mechanism, design and implementation of node management, auxiliary analysis, instruction management and condition monitoring are analyzed. Multi-dimensional multi-source heterogeneous fault data are used as training samples to achieve automatic detection of equipment operating conditions. Reinforcement learning theory is applied to equipment fault diagnosis to achieve automatic judgment of test data, so as to establish the active early warning mechanism of equipment.","PeriodicalId":518671,"journal":{"name":"2024 IEEE 4th International Conference on Power, Electronics and Computer Applications (ICPECA)","volume":"38 1","pages":"1043-1047"},"PeriodicalIF":0.0,"publicationDate":"2024-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530504","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-26DOI: 10.1109/ICPECA60615.2024.10471046
Jingyuan Bai
With the continuous improvement of big data and computing power, deep learning models have achieved remarkable results in the field of image recognition, but building and training a deep neural network from scratch often requires a large amount of annotated data and expensive computing resources. This article first outlines the basic principles and challenges of deep learning in image classification tasks. Especially for task scenarios with small samples or scarce annotations, traditional deep learning models are prone to overfitting and insufficient generalization performance. Transfer learning is introduced into this study as an important strategy. Through deep models (such as ResNet, VGG, etc.) pre-trained on large-scale image data sets (such as ImageNet), universal feature representations are extracted. And we transfer these pre-trained model parameters to specific target image classification tasks for fine-tuning. Furthermore, this article elaborates on several typical applications of transfer learning in deep learning models, and analyzes how transfer learning can effectively help improve the accuracy of image classification on the target data set based on actual cases. The experimental part compares the results of directly training a new model and using the transfer learning method to initialize the model and then train on a variety of target data sets. The experiment proves that transfer learning can significantly improve the learning efficiency and final classification performance of the model under limited samples.
{"title":"Research and Application of Deep Learning Based on Transfer Learning in Image Classification Tasks","authors":"Jingyuan Bai","doi":"10.1109/ICPECA60615.2024.10471046","DOIUrl":"https://doi.org/10.1109/ICPECA60615.2024.10471046","url":null,"abstract":"With the continuous improvement of big data and computing power, deep learning models have achieved remarkable results in the field of image recognition, but building and training a deep neural network from scratch often requires a large amount of annotated data and expensive computing resources. This article first outlines the basic principles and challenges of deep learning in image classification tasks. Especially for task scenarios with small samples or scarce annotations, traditional deep learning models are prone to overfitting and insufficient generalization performance. Transfer learning is introduced into this study as an important strategy. Through deep models (such as ResNet, VGG, etc.) pre-trained on large-scale image data sets (such as ImageNet), universal feature representations are extracted. And we transfer these pre-trained model parameters to specific target image classification tasks for fine-tuning. Furthermore, this article elaborates on several typical applications of transfer learning in deep learning models, and analyzes how transfer learning can effectively help improve the accuracy of image classification on the target data set based on actual cases. The experimental part compares the results of directly training a new model and using the transfer learning method to initialize the model and then train on a variety of target data sets. The experiment proves that transfer learning can significantly improve the learning efficiency and final classification performance of the model under limited samples.","PeriodicalId":518671,"journal":{"name":"2024 IEEE 4th International Conference on Power, Electronics and Computer Applications (ICPECA)","volume":"4 3","pages":"1292-1297"},"PeriodicalIF":0.0,"publicationDate":"2024-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530511","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2024-01-26DOI: 10.1109/ICPECA60615.2024.10470977
Siyuan Shen
This article introduces an improvement in the YOLOv5 architecture by incorporating the CBAM (Convolutional Block Attention Module) attention module at the neck network end. CBAM is added after each concatenation operation to enhance the focus on small targets and optimize the fusion features in the neck. The role of CBAM is to strengthen the extraction of features by automatically ignoring irrelevant information, focusing on the fusion of crucial features, thereby improving the model's analytical capabilities for complex scenes. Experimental results indicate that the addition of the CBAM module successfully enhances the YOLOv5s model by highlighting key features and suppressing unimportant ones. This results in output feature maps containing more valuable information, significantly improving the accuracy of object detection. This improvement has shown positive effects in small object detection, feature fusion, and model speed.
{"title":"Research on Small Object Detection Algorithm Based on YOLOv5","authors":"Siyuan Shen","doi":"10.1109/ICPECA60615.2024.10470977","DOIUrl":"https://doi.org/10.1109/ICPECA60615.2024.10470977","url":null,"abstract":"This article introduces an improvement in the YOLOv5 architecture by incorporating the CBAM (Convolutional Block Attention Module) attention module at the neck network end. CBAM is added after each concatenation operation to enhance the focus on small targets and optimize the fusion features in the neck. The role of CBAM is to strengthen the extraction of features by automatically ignoring irrelevant information, focusing on the fusion of crucial features, thereby improving the model's analytical capabilities for complex scenes. Experimental results indicate that the addition of the CBAM module successfully enhances the YOLOv5s model by highlighting key features and suppressing unimportant ones. This results in output feature maps containing more valuable information, significantly improving the accuracy of object detection. This improvement has shown positive effects in small object detection, feature fusion, and model speed.","PeriodicalId":518671,"journal":{"name":"2024 IEEE 4th International Conference on Power, Electronics and Computer Applications (ICPECA)","volume":"61 2","pages":"937-942"},"PeriodicalIF":0.0,"publicationDate":"2024-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"140530299","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：