This paper proposes an end-to-end first-stage pallet detection algorithm with short training time and high detection accuracy based on the pre-detection staged material detection algorithm. Not only can it detect known materials, blank areas and fixed material areas on pallets, but also unknown and unwanted materials that are mixed and misplaced on pallets on automated assembly lines. It employs ResNet18 as the backbone network, incorporates the Convolutional Block Attention Module (CBAM) to improve model stability and accuracy, and optimizes the detection model using the multitask loss function based on Complete-IoU(CIoU) and cross entropy. The experimental results show that when compared to the original phased detection algorithm using YOLOv5s trained on four NVIDIA GeForce RTX 2080 Ti for 18 h, the phased detection algorithm used in this study's first stage material detection algorithm achieves 98% overall recognition accuracy, which is 7% higher than the original phased algorithm (91%). It also greatly reduces the model training time and allows rapid model deployment.
{"title":"Detection of Material on a Tray in an Automatic Assembly Line Based on Convolution Attention and Multitask Loss","authors":"Dunli Hu, Yuting Zhang, Xiaoping Zhang, Xiangdong Zhang","doi":"10.1109/ICCR55715.2022.10053928","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053928","url":null,"abstract":"This paper proposes an end-to-end first-stage pallet detection algorithm with short training time and high detection accuracy based on the pre-detection staged material detection algorithm. Not only can it detect known materials, blank areas and fixed material areas on pallets, but also unknown and unwanted materials that are mixed and misplaced on pallets on automated assembly lines. It employs ResNet18 as the backbone network, incorporates the Convolutional Block Attention Module (CBAM) to improve model stability and accuracy, and optimizes the detection model using the multitask loss function based on Complete-IoU(CIoU) and cross entropy. The experimental results show that when compared to the original phased detection algorithm using YOLOv5s trained on four NVIDIA GeForce RTX 2080 Ti for 18 h, the phased detection algorithm used in this study's first stage material detection algorithm achieves 98% overall recognition accuracy, which is 7% higher than the original phased algorithm (91%). It also greatly reduces the model training time and allows rapid model deployment.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"13 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125122012","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 : 2022-12-02DOI: 10.1109/ICCR55715.2022.10053865
Beixian Lai, Zhiwen Li, Weibing Li, Yongping Pan
Visual servoing can effectively control robots using visual feedback information to improve the intelligence and reliability. In most existing dynamics-based image-based visual servoing methods, a restricted condition that the number of the feature points is no larger than 3 is needed to achieve pixel error convergence, which makes them difficlut to achieve three-dimensional (3-D) pose control since at least 4 feature points on a plane are needed to determine the unique end-effector pose. This paper puts forward to a dynamics-based adaptive homography-based visual servoing (HBVS) controller to regulate robot manipulators with eye-in-hand monocular cameras to the desired pose under unknown but constant feature positions. The uncertain depth is represented as a linear form of its position parameters in the Cartesian space, and a composite learning technique is applied to guarantee parameter convergence under a much weaker condition of interval excitation than persistent excitation, resulting in exact depth estimation and 3-D pose regulation. Experiments on a collaborative robot with 7 degrees of freedom named Franka Emika Panda have illustrated the effectiveness of the proposed method.
{"title":"Homography-Based Visual Servoing for Eye-in-Hand Robots with Unknown Feature Positions","authors":"Beixian Lai, Zhiwen Li, Weibing Li, Yongping Pan","doi":"10.1109/ICCR55715.2022.10053865","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053865","url":null,"abstract":"Visual servoing can effectively control robots using visual feedback information to improve the intelligence and reliability. In most existing dynamics-based image-based visual servoing methods, a restricted condition that the number of the feature points is no larger than 3 is needed to achieve pixel error convergence, which makes them difficlut to achieve three-dimensional (3-D) pose control since at least 4 feature points on a plane are needed to determine the unique end-effector pose. This paper puts forward to a dynamics-based adaptive homography-based visual servoing (HBVS) controller to regulate robot manipulators with eye-in-hand monocular cameras to the desired pose under unknown but constant feature positions. The uncertain depth is represented as a linear form of its position parameters in the Cartesian space, and a composite learning technique is applied to guarantee parameter convergence under a much weaker condition of interval excitation than persistent excitation, resulting in exact depth estimation and 3-D pose regulation. Experiments on a collaborative robot with 7 degrees of freedom named Franka Emika Panda have illustrated the effectiveness of the proposed method.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"27 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125509056","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 : 2022-12-02DOI: 10.1109/ICCR55715.2022.10053883
Hao Wang, Zhiru Chen, Jun Wang, Lijun Lu, Mingzhe Li
To achieve the consensus for discrete-time multi-agent systems, an optimal control policy is designed based on off-policy reinforcement learning. By utilizing centralized learning and decentralized execution, we first define a centralized and shared value function. Then, a value iteration adaptive dynamic programming method is proposed to approach the solution of the Bellman optimality equation with convergence analysis. Furthermore, the actor-critic structure is given for the implementation purpose, where one single-critic network is given to approach the optimal centralized value function, and multi-actor networks are decentralized based on the local observation from the neighbors to obtain the optimal policy for each agent. Finally, the proposed algorithm is verified in a leader-follower consensus case.
{"title":"Optimal Control for Multi-agent Systems Using Off-Policy Reinforcement Learning","authors":"Hao Wang, Zhiru Chen, Jun Wang, Lijun Lu, Mingzhe Li","doi":"10.1109/ICCR55715.2022.10053883","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053883","url":null,"abstract":"To achieve the consensus for discrete-time multi-agent systems, an optimal control policy is designed based on off-policy reinforcement learning. By utilizing centralized learning and decentralized execution, we first define a centralized and shared value function. Then, a value iteration adaptive dynamic programming method is proposed to approach the solution of the Bellman optimality equation with convergence analysis. Furthermore, the actor-critic structure is given for the implementation purpose, where one single-critic network is given to approach the optimal centralized value function, and multi-actor networks are decentralized based on the local observation from the neighbors to obtain the optimal policy for each agent. Finally, the proposed algorithm is verified in a leader-follower consensus case.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"5 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122457861","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 : 2022-12-02DOI: 10.1109/ICCR55715.2022.10053875
Jiawei Tian, Dazi Li
The current path planning algorithms such as A-star(all stars) algorithm and RRT (Rapidly-exploring Random Trees) algorithm can meet the obstacle avoidance planning of the 6-DOF robot, but the smoothness of the path is not considered. Working in an unreasonable path for a long time will produce a great load on the joints of the 6-DOF robot and seriously affect its life. In this paper, we use reinforcement learning reconcile A-star algorithm and RRT algorithm for smooth path planning of the robot. Experimental results show that compared with A-star algorithm and RRT algorithm, the fusion algorithm has smoother path and more reasonable time.
目前的路径规划算法如A-star(all stars)算法和RRT (rapid -exploring Random Trees)算法可以满足六自由度机器人的避障规划，但没有考虑路径的平滑性。长时间在不合理的路径上工作，会对六自由度机器人的关节产生很大的载荷，严重影响其寿命。在本文中，我们使用强化学习调和A-star算法和RRT算法来进行机器人的平滑路径规划。实验结果表明，与A-star算法和RRT算法相比，该融合算法路径更平滑，时间更合理。
{"title":"Smooth Path Planning of 6-DOF Robot Based on Reinforcement Learning","authors":"Jiawei Tian, Dazi Li","doi":"10.1109/ICCR55715.2022.10053875","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053875","url":null,"abstract":"The current path planning algorithms such as A-star(all stars) algorithm and RRT (Rapidly-exploring Random Trees) algorithm can meet the obstacle avoidance planning of the 6-DOF robot, but the smoothness of the path is not considered. Working in an unreasonable path for a long time will produce a great load on the joints of the 6-DOF robot and seriously affect its life. In this paper, we use reinforcement learning reconcile A-star algorithm and RRT algorithm for smooth path planning of the robot. Experimental results show that compared with A-star algorithm and RRT algorithm, the fusion algorithm has smoother path and more reasonable time.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128259165","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 : 2022-12-02DOI: 10.1109/ICCR55715.2022.10053861
Guorun Yang, Y. Qiao, Jianping Shit, Zhe Wang
The long-tailed object distribution poses great challenges for autonomous driving. And the field collection of long-tailed objects is difficult and high-cost. In this paper, we propose a novel data mining approach for those long-tailed objects. The softmax distribution produced by CLIP model is adopted as the representation of cropped objects in the image. Then for each long-tailed classification, the category grouping is performed to divide the text concepts into three sets. Finally, combining the softmax representation with the grouped categories, we develop an effective softmax mining algorithm to search and identify the long-tailed objects from the large database. Experiments demonstrate that the proposed method outperforms the baseline results and accurately finds the long-tailed data.
{"title":"Long-Tailed Object Mining Based on CLIP Model for Autonomous Driving","authors":"Guorun Yang, Y. Qiao, Jianping Shit, Zhe Wang","doi":"10.1109/ICCR55715.2022.10053861","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053861","url":null,"abstract":"The long-tailed object distribution poses great challenges for autonomous driving. And the field collection of long-tailed objects is difficult and high-cost. In this paper, we propose a novel data mining approach for those long-tailed objects. The softmax distribution produced by CLIP model is adopted as the representation of cropped objects in the image. Then for each long-tailed classification, the category grouping is performed to divide the text concepts into three sets. Finally, combining the softmax representation with the grouped categories, we develop an effective softmax mining algorithm to search and identify the long-tailed objects from the large database. Experiments demonstrate that the proposed method outperforms the baseline results and accurately finds the long-tailed data.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"38 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114236908","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 : 2022-12-02DOI: 10.1109/ICCR55715.2022.10053856
Wen Qi, Haoyu Fan, Yancai Xu, Hang Su, A. Aliverti
Finger gesture recognition using surface electromyography (sEMG) became an efficient Human-Robot Interaction (HRI) solution. Although Machine Learning (ML) techniques are widely applied in this field, the general solutions for labeling and collecting big datasets impose time-consuming implementation and heavy workloads. In this paper, a new deep learning structure, namely three-dimensional convolutional long short-term memory neural networks (3D-CLDNN) for finger gesture identification based on depth vision and sEMG signals, was proposed for human-machine interaction. It automatically labels the depth data by the self-organizing map (SOM) and predicts the hand gesture only adopting sEMG signals. The 3D-CLDNN method is integrated to improve the recognition rate and computational speed. The results showed the highest clustering accuracy (98.60%) and highest accuracy (84.40%) with the lowest computational time compared with different approaches. Finally, real-time human-machine interaction experiments are performed to demonstrate its efficiency.
{"title":"A 3D-CLDNN Based Multiple Data Fusion Framework for Finger Gesture Recognition in Human-Robot Interaction","authors":"Wen Qi, Haoyu Fan, Yancai Xu, Hang Su, A. Aliverti","doi":"10.1109/ICCR55715.2022.10053856","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053856","url":null,"abstract":"Finger gesture recognition using surface electromyography (sEMG) became an efficient Human-Robot Interaction (HRI) solution. Although Machine Learning (ML) techniques are widely applied in this field, the general solutions for labeling and collecting big datasets impose time-consuming implementation and heavy workloads. In this paper, a new deep learning structure, namely three-dimensional convolutional long short-term memory neural networks (3D-CLDNN) for finger gesture identification based on depth vision and sEMG signals, was proposed for human-machine interaction. It automatically labels the depth data by the self-organizing map (SOM) and predicts the hand gesture only adopting sEMG signals. The 3D-CLDNN method is integrated to improve the recognition rate and computational speed. The results showed the highest clustering accuracy (98.60%) and highest accuracy (84.40%) with the lowest computational time compared with different approaches. Finally, real-time human-machine interaction experiments are performed to demonstrate its efficiency.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"10 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127891951","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 : 2022-12-02DOI: 10.1109/ICCR55715.2022.10053876
Jing Luo, Yu Li, Xiaoli Liu, Jianwen Hu, Bo Wang
As a representation of teleoperated robots, surgical robots based on teleoperation are widely applied in the medical field. Generally, it can greatly guarantee the performance of microsurgery in terms of control and postoperative recovery by using a surgical robot in comparison with traditional surgical operation. However, the performance of surgical robots is greatly disturbed by the physiological tremor of surgeon in the process of operation. In order to cancel the impacts of tremor signal, a neural-network-based (NN-based) algorithm is developed in this paper. For the proposed NN-based approach, we develop a hybrid wavelet basis function to deal with the variable tremor signal. Additionally, the proposed method can cancel the tremor signals based on the excellent ability of nonlinear mapping and generalization and does not rely on a priori structural parameters. In order to evaluate the performance of the proposed method, comparative experiments of five different kinds of NN-based tremor filter are performed by using tremor signals with different frequencies and amplitudes. Experimental results validated that the proposed algorithm can achieve the performance of suppressing the tremor signal of the processing error. It is can be noted that the surgical robots can ensure the control performance of the surgical robots by using the developed NN-based filter. The developed method can also be applied as a filter for suppressing vibrations of processing operations in the future, such as chatter in micro-milling.
{"title":"Neural-network-based Algorithm for Cancelling Tremor in Surgical Robots","authors":"Jing Luo, Yu Li, Xiaoli Liu, Jianwen Hu, Bo Wang","doi":"10.1109/ICCR55715.2022.10053876","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053876","url":null,"abstract":"As a representation of teleoperated robots, surgical robots based on teleoperation are widely applied in the medical field. Generally, it can greatly guarantee the performance of microsurgery in terms of control and postoperative recovery by using a surgical robot in comparison with traditional surgical operation. However, the performance of surgical robots is greatly disturbed by the physiological tremor of surgeon in the process of operation. In order to cancel the impacts of tremor signal, a neural-network-based (NN-based) algorithm is developed in this paper. For the proposed NN-based approach, we develop a hybrid wavelet basis function to deal with the variable tremor signal. Additionally, the proposed method can cancel the tremor signals based on the excellent ability of nonlinear mapping and generalization and does not rely on a priori structural parameters. In order to evaluate the performance of the proposed method, comparative experiments of five different kinds of NN-based tremor filter are performed by using tremor signals with different frequencies and amplitudes. Experimental results validated that the proposed algorithm can achieve the performance of suppressing the tremor signal of the processing error. It is can be noted that the surgical robots can ensure the control performance of the surgical robots by using the developed NN-based filter. The developed method can also be applied as a filter for suppressing vibrations of processing operations in the future, such as chatter in micro-milling.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"5 7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"134226259","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 : 2022-12-02DOI: 10.1109/ICCR55715.2022.10053920
Songyi Xiao, Bo Zhao, Derong Liu
Neural architecture search (NAS) has low efficiency in evaluating a large number of candidate architectures. As an efficient evaluation method, accuracy predictor-based NAS algorithms have become popular because the performance (accuracy) can be evaluated without training the candidate architectures. However, accuracy predictors still need some evaluated architectures that are difficult to train for achieving promising performance. In order to break this bottleneck, we investigate a semi-supervised accuracy predictor-based evolutionary NAS method (MSNAS) which requires only a small number of evaluated neural architectures. The accuracy predictor obtains high prediction performance by extracting the evaluated architectures, strong regressors and truncation mechanism. To find truly high-accuracy candidate architectures more easily, the multi-objective optimization method is presented to trade-off the prediction accuracy and confidence of candidate architectures. The MSNAS variants from different strong regressors are employed to validate the competitive performance of the MSNAS on NAS-Bench 201.
{"title":"Evolutionary Neural Architecture Search with Semi-supervised Accuracy Predictor","authors":"Songyi Xiao, Bo Zhao, Derong Liu","doi":"10.1109/ICCR55715.2022.10053920","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053920","url":null,"abstract":"Neural architecture search (NAS) has low efficiency in evaluating a large number of candidate architectures. As an efficient evaluation method, accuracy predictor-based NAS algorithms have become popular because the performance (accuracy) can be evaluated without training the candidate architectures. However, accuracy predictors still need some evaluated architectures that are difficult to train for achieving promising performance. In order to break this bottleneck, we investigate a semi-supervised accuracy predictor-based evolutionary NAS method (MSNAS) which requires only a small number of evaluated neural architectures. The accuracy predictor obtains high prediction performance by extracting the evaluated architectures, strong regressors and truncation mechanism. To find truly high-accuracy candidate architectures more easily, the multi-objective optimization method is presented to trade-off the prediction accuracy and confidence of candidate architectures. The MSNAS variants from different strong regressors are employed to validate the competitive performance of the MSNAS on NAS-Bench 201.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"8 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133267798","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 : 2022-12-02DOI: 10.1109/ICCR55715.2022.10053868
Xiao-Zhi Gao
In this paper, a sliding mode control (SMC) based on nonlinear disturbance observer and intermittent control is proposed to maximize the security of cyber-physical systems (CPSs), aiming at the cyber-attacks and physical uncertainties of cyber-physical systems. In the CPSs, the transmission of information data and control signals to the remote end through the network may lead to cyber attacks, and there will be uncertainties in the physical system. Therefore, this paper establishes a CPSs model that includes network attacks and physical uncertainties. Secondly, according to the analysis of the mathematical model, an adaptive SMC based on disturbance observer and intermittent control is designed to keep the CPSs stable in the presence of network attacks and physical uncertainties. In this strategy, the adaptive strategy suppresses the controller The chattering of the output. Intermittent control breaks the limitations of traditional continuous control to ensure efficient use of resources. Finally, to prove the control performance of the controller, numerical simulation results are given.
{"title":"Sliding Mode Control Based on Disturbance Observer for Cyber-Physical Systems Security","authors":"Xiao-Zhi Gao","doi":"10.1109/ICCR55715.2022.10053868","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053868","url":null,"abstract":"In this paper, a sliding mode control (SMC) based on nonlinear disturbance observer and intermittent control is proposed to maximize the security of cyber-physical systems (CPSs), aiming at the cyber-attacks and physical uncertainties of cyber-physical systems. In the CPSs, the transmission of information data and control signals to the remote end through the network may lead to cyber attacks, and there will be uncertainties in the physical system. Therefore, this paper establishes a CPSs model that includes network attacks and physical uncertainties. Secondly, according to the analysis of the mathematical model, an adaptive SMC based on disturbance observer and intermittent control is designed to keep the CPSs stable in the presence of network attacks and physical uncertainties. In this strategy, the adaptive strategy suppresses the controller The chattering of the output. Intermittent control breaks the limitations of traditional continuous control to ensure efficient use of resources. Finally, to prove the control performance of the controller, numerical simulation results are given.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"15 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133141033","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 : 2022-12-02DOI: 10.1109/ICCR55715.2022.10053869
Ziyang Yu, Dongsheng Yang, Weirong Wu, Yingchun Wang, Yanhong Luo
Due to low resolution or few features, small object detection has become a difficult problem in the field of image recognition. This paper proposes a fast convergence detection algorithm for small image objects based on distance intersection over union. First of all, EnlightenGAN is used to enhance the image, reduce image noise, and highlight the detection object features. Then, a loss function design of YOLOv5 network based on distance intersection over union is proposed. This method speeds up the gradient regression of the network, greatly shortens the training time of the YOLOv5 network, and improves the detection accuracy. The experimental results using the WiderPerson dataset and the VOC07++12 dataset show that, compared with the traditional YOLOv5 network image detection results, the method proposed in this paper improves AP0.5 by 4.4% and 3.3%, and APs by 6.8% and 2.5%, respectively, which verifies the effectiveness of this method.
{"title":"Fast Convergence Detection Algorithm of Image Small Object Based on Distance Intersection over Union","authors":"Ziyang Yu, Dongsheng Yang, Weirong Wu, Yingchun Wang, Yanhong Luo","doi":"10.1109/ICCR55715.2022.10053869","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053869","url":null,"abstract":"Due to low resolution or few features, small object detection has become a difficult problem in the field of image recognition. This paper proposes a fast convergence detection algorithm for small image objects based on distance intersection over union. First of all, EnlightenGAN is used to enhance the image, reduce image noise, and highlight the detection object features. Then, a loss function design of YOLOv5 network based on distance intersection over union is proposed. This method speeds up the gradient regression of the network, greatly shortens the training time of the YOLOv5 network, and improves the detection accuracy. The experimental results using the WiderPerson dataset and the VOC07++12 dataset show that, compared with the traditional YOLOv5 network image detection results, the method proposed in this paper improves AP0.5 by 4.4% and 3.3%, and APs by 6.8% and 2.5%, respectively, which verifies the effectiveness of this method.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"6 1-2 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-12-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124043454","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：