Pub Date : 2022-12-02DOI: 10.1109/ICCR55715.2022.10053854
Yang Liu, Jiajun Zhang, W. Liao
Wearable lower limb rehabilitation exoskeleton robots play a positive role in lower limb rehabilitation training and assistance walking for patients with lower limb disorders. Firstly, the 3 degrees of freedom link-based dynamic model with friction is established by the Lagrange method. Secondly, a parameter identification experiment is designed based on a lower limb exoskeleton prototype. It contains three parts: static experiment of discrete controlled by specified position, dynamic experiment of uniform speed motion controlled by linear excitations, and dynamic experiment of continuous motion controlled by sinusoidal excitations. During the process of experiment, several terms in joint output torque expression are set to zero for simplicity of calculation, and leave the parameters to be identified. Furthermore, based on the acquired actuator torque data, nine parameters are identified by plotting and curves fitting with the least square method, including inertial parameters, static friction and Coulomb viscous friction. Finally, the parameter identification results are verified through comparing the torque measured by experiment and estimated by model.
{"title":"Dynamic Modeling and Identification of Wearable Lower Limb Rehabilitation Exoskeleton Robots","authors":"Yang Liu, Jiajun Zhang, W. Liao","doi":"10.1109/ICCR55715.2022.10053854","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053854","url":null,"abstract":"Wearable lower limb rehabilitation exoskeleton robots play a positive role in lower limb rehabilitation training and assistance walking for patients with lower limb disorders. Firstly, the 3 degrees of freedom link-based dynamic model with friction is established by the Lagrange method. Secondly, a parameter identification experiment is designed based on a lower limb exoskeleton prototype. It contains three parts: static experiment of discrete controlled by specified position, dynamic experiment of uniform speed motion controlled by linear excitations, and dynamic experiment of continuous motion controlled by sinusoidal excitations. During the process of experiment, several terms in joint output torque expression are set to zero for simplicity of calculation, and leave the parameters to be identified. Furthermore, based on the acquired actuator torque data, nine parameters are identified by plotting and curves fitting with the least square method, including inertial parameters, static friction and Coulomb viscous friction. Finally, the parameter identification results are verified through comparing the torque measured by experiment and estimated by model.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"67 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":"122545601","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/ICCR55715.2022.10053888
H. Huang, Ding Wang, Junlong Wu, Lingzhi Hu
This paper develops a novel value iteration (VI) scheme and an online VI algorithm, to address the discounted optimal control problems of affine discrete-time nonlinear systems. First, we provide the derivation of the novel VI. Second, we analyze the convergence and monotonicity of the iterative value function sequence, as well as the admissibility of the iterative control. Third, based on the theory of the attraction domain and the novel VI scheme, an online VI algorithm is proposed to implement the stability analysis of the controlled system. It is worth noting that the current control during the online control stage is determined by the location of the current state. Finally, a simulation example is involved to demonstrate the performance of the developed algorithms.
{"title":"Innovative Discounted Optimal Control Design via Offline and Online Formulations for Affine Systems","authors":"H. Huang, Ding Wang, Junlong Wu, Lingzhi Hu","doi":"10.1109/ICCR55715.2022.10053888","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053888","url":null,"abstract":"This paper develops a novel value iteration (VI) scheme and an online VI algorithm, to address the discounted optimal control problems of affine discrete-time nonlinear systems. First, we provide the derivation of the novel VI. Second, we analyze the convergence and monotonicity of the iterative value function sequence, as well as the admissibility of the iterative control. Third, based on the theory of the attraction domain and the novel VI scheme, an online VI algorithm is proposed to implement the stability analysis of the controlled system. It is worth noting that the current control during the online control stage is determined by the location of the current state. Finally, a simulation example is involved to demonstrate the performance of the developed algorithms.","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":"117064615","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/ICCR55715.2022.10053890
Jiangfeng Liu, Yongbin Guo, Jinbiao Chen, Zixu Wang, Aihua Mao
We propose a cross-lingual TTS model based on the neural network. The model is capable of synthesizing speech across languages and translating the speaker's timbre. It uses a few seconds of untranscribed reference audio of the target speaker to synthesize the new speech of that speaker. The model consists of a separate speaker encoder, STT Translator, synthesizer, and vocoder. We decouple speaker information and speech to build a speaker recognition network. Our synthesizer is mainly built based on the Tacotron model and is divided into three parts: encoder, attention mechanism and decoder. The vocoder, on the other hand, is based on two methods, WaveRNN and HiFi-GAN, and serves to predict the synthesized waveform using the Mel spectrum. We conducted experiments to analyze the effectiveness of our approach. Besides, we also analyzed the effect of different datasets on the training effect.
{"title":"Speech Synthesis for Speaker Timbre Translation Across Languages","authors":"Jiangfeng Liu, Yongbin Guo, Jinbiao Chen, Zixu Wang, Aihua Mao","doi":"10.1109/ICCR55715.2022.10053890","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053890","url":null,"abstract":"We propose a cross-lingual TTS model based on the neural network. The model is capable of synthesizing speech across languages and translating the speaker's timbre. It uses a few seconds of untranscribed reference audio of the target speaker to synthesize the new speech of that speaker. The model consists of a separate speaker encoder, STT Translator, synthesizer, and vocoder. We decouple speaker information and speech to build a speaker recognition network. Our synthesizer is mainly built based on the Tacotron model and is divided into three parts: encoder, attention mechanism and decoder. The vocoder, on the other hand, is based on two methods, WaveRNN and HiFi-GAN, and serves to predict the synthesized waveform using the Mel spectrum. We conducted experiments to analyze the effectiveness of our approach. Besides, we also analyzed the effect of different datasets on the training effect.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"16 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":"123388826","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/ICCR55715.2022.10053924
Shi Liu, Tehuan Chen, Chao Xu
Continuous stirred tank reactor (CSTR) is one of the most common industrial equipment in petroleum and chemical industry, and is widely used in regrouping, fermentation engineering and additive preparation. In general, CSTR is used to prepare a fixed concentration of the output product. Accurate and fast monitoring of the changes in the state quantities of the CSTR chemical reaction process becomes the most important aspect before implementing excellent control. This paper presents a neural network observer with a residual network as the core component. In addition, the operations of the neural network are also matrixed to isolate the nonlinearities as much as possible. Finally we conduct numerical experiments in MATLAB R2018b based on SIMULINK framework to verify the feasibility of our strategy.
{"title":"A CSTR State Observer Based on Residual Neural Network","authors":"Shi Liu, Tehuan Chen, Chao Xu","doi":"10.1109/ICCR55715.2022.10053924","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053924","url":null,"abstract":"Continuous stirred tank reactor (CSTR) is one of the most common industrial equipment in petroleum and chemical industry, and is widely used in regrouping, fermentation engineering and additive preparation. In general, CSTR is used to prepare a fixed concentration of the output product. Accurate and fast monitoring of the changes in the state quantities of the CSTR chemical reaction process becomes the most important aspect before implementing excellent control. This paper presents a neural network observer with a residual network as the core component. In addition, the operations of the neural network are also matrixed to isolate the nonlinearities as much as possible. Finally we conduct numerical experiments in MATLAB R2018b based on SIMULINK framework to verify the feasibility of our strategy.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"122 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":"123575239","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/ICCR55715.2022.10053864
Yansui Song, Shuai Liang, Erzhuo Niu, Bin Xu
This paper investigates the trajectory optimization and tracking control for the perch maneuver of a fixed-wing unmanned aerial vehicle (UAV). An important aspect of the perch maneuver is that it provides a fast landing for UAVs on fixed points, which could be useful to solve the problem of landing dornes on warship or in tight areas. Optimal trajectory optimization is one of the main concerns of the technology, which is optimised for the shortest trajectory length and minimal energy consumption of the actuator in this paper. In addition, high-precision trajectory tracking control is required, but it is difficult due to the contradiction between variable model parameters and high-precision trajectory tracking control at high angles of attack flight. Toward this end, we developed a cascade incremental nonlinear dynamic inverse (INDI) controller which has a great robustness to the model uncertainties. As a result of simulation, it is verified that the INDI controller can maintain high trajectory tracking accuracy even at a large model deviation, and that it has a better control performance than a linear quadratic controller.
{"title":"A Perched Landing Control Method Based on Incremental Nonlinear Dynamic Inverse","authors":"Yansui Song, Shuai Liang, Erzhuo Niu, Bin Xu","doi":"10.1109/ICCR55715.2022.10053864","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053864","url":null,"abstract":"This paper investigates the trajectory optimization and tracking control for the perch maneuver of a fixed-wing unmanned aerial vehicle (UAV). An important aspect of the perch maneuver is that it provides a fast landing for UAVs on fixed points, which could be useful to solve the problem of landing dornes on warship or in tight areas. Optimal trajectory optimization is one of the main concerns of the technology, which is optimised for the shortest trajectory length and minimal energy consumption of the actuator in this paper. In addition, high-precision trajectory tracking control is required, but it is difficult due to the contradiction between variable model parameters and high-precision trajectory tracking control at high angles of attack flight. Toward this end, we developed a cascade incremental nonlinear dynamic inverse (INDI) controller which has a great robustness to the model uncertainties. As a result of simulation, it is verified that the INDI controller can maintain high trajectory tracking accuracy even at a large model deviation, and that it has a better control performance than a linear quadratic controller.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"28 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":"121611975","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/ICCR55715.2022.10053881
Yuge Xu, Zixing Guo, Xie Zhang, Chuanlong Lv
Aluminum profiles are widely used in many industries with good characteristics. Surface defects of aluminum profiles will affect the quality, reliability and safety of products. In recent years, deep learning has been applied in aluminum profile surface defect detection. However, there are still some problems unsolved. The tiny and narrow defects are easily ignored in feature extraction. The length-width ratio of different aluminum surface defects varies widely, but the fixed anchor boxes in traditional deep learning algorithms will easily miss defects. Due to the lack of training on background images, some backgrounds are easily misidentified as defects. To address these problems, a novel Cascade R-CNN network with deformable convolution, guided anchoring and sample augmentation (GAE-Cascade R-CNN model) is proposed. The deformable convolution enhances the feature extraction ability of the network. The guided anchoring reduces the missed detection by automatically generating anchors to match narrow defects. The sample augmentation effectively reduces missed defect detection by training a large number of background images. The experimental results show that the proposed GAE-Cascade R-CNN model can achieve accuracy of 98.85% for identification and mean average precision (mAP) of 80.55% for surface defect detection of aluminum profiles. The performance of the proposed network outperforms other deep learning methods in terms of both missed detection rate and false detection rate.
{"title":"Research on Surface Defect Detection of Aluminum Based on Improved Cascade R-CNN","authors":"Yuge Xu, Zixing Guo, Xie Zhang, Chuanlong Lv","doi":"10.1109/ICCR55715.2022.10053881","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053881","url":null,"abstract":"Aluminum profiles are widely used in many industries with good characteristics. Surface defects of aluminum profiles will affect the quality, reliability and safety of products. In recent years, deep learning has been applied in aluminum profile surface defect detection. However, there are still some problems unsolved. The tiny and narrow defects are easily ignored in feature extraction. The length-width ratio of different aluminum surface defects varies widely, but the fixed anchor boxes in traditional deep learning algorithms will easily miss defects. Due to the lack of training on background images, some backgrounds are easily misidentified as defects. To address these problems, a novel Cascade R-CNN network with deformable convolution, guided anchoring and sample augmentation (GAE-Cascade R-CNN model) is proposed. The deformable convolution enhances the feature extraction ability of the network. The guided anchoring reduces the missed detection by automatically generating anchors to match narrow defects. The sample augmentation effectively reduces missed defect detection by training a large number of background images. The experimental results show that the proposed GAE-Cascade R-CNN model can achieve accuracy of 98.85% for identification and mean average precision (mAP) of 80.55% for surface defect detection of aluminum profiles. The performance of the proposed network outperforms other deep learning methods in terms of both missed detection rate and false detection rate.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"58 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":"127624495","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/ICCR55715.2022.10053859
Hanjun Xie, Qin-ruo Wang
Aiming at the dynamic coupling problem of the dual-drive motors caused by the position change of heavy-load working head on gantry stage, this paper establishes an accurate electromechanical mathematical model for rigid-flexible coupling characteristics of the platform based on first principles. The model includes the crossbeam's linear motion and rotational motion of the non-constant moment of inertia ${J}$, and the cross-coupling force between the dual drive motors is quantified by defining the virtual centroid of the crossbeam. Finally, the effectiveness of the model is verified by the frequency response experiment of a actual system.
{"title":"Coupling Modeling Analysis of Synchronous Direct-Drive Gantry Laser Cutting Stage with Heavy-Load","authors":"Hanjun Xie, Qin-ruo Wang","doi":"10.1109/ICCR55715.2022.10053859","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053859","url":null,"abstract":"Aiming at the dynamic coupling problem of the dual-drive motors caused by the position change of heavy-load working head on gantry stage, this paper establishes an accurate electromechanical mathematical model for rigid-flexible coupling characteristics of the platform based on first principles. The model includes the crossbeam's linear motion and rotational motion of the non-constant moment of inertia ${J}$, and the cross-coupling force between the dual drive motors is quantified by defining the virtual centroid of the crossbeam. Finally, the effectiveness of the model is verified by the frequency response experiment of a actual system.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"45 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":"124966912","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/ICCR55715.2022.10053849
Xiaoxiao Li, Yinhe Wang, Shengping Li
In this research, the model following adaptive control problem for the complex dynamical networks with adaptive coupling is investigated. Firstly, from the large-scale system perspective, the complex dynamical networks studied in this paper is composed of many nodes coupled with each other, where each node is a dynamic basic unit with detailed content, which is modeled by the vector differential equation. Next, based on the Lyapunov stability theory, an appropriate adaptive control scheme and the adaptive coupling are designed for the controller plant, so that the controlled plant can asymptotically track its model following target. Furthermore, each node has a different model following target, which is actually achieve multi-target tracking control. Finally, numerical simulation is given to verify the effectiveness and correctness of the control scheme proposed in this paper.
{"title":"Model Following Adaptive Control of Complex Dynamical Networks with the Adaptive Coupling","authors":"Xiaoxiao Li, Yinhe Wang, Shengping Li","doi":"10.1109/ICCR55715.2022.10053849","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053849","url":null,"abstract":"In this research, the model following adaptive control problem for the complex dynamical networks with adaptive coupling is investigated. Firstly, from the large-scale system perspective, the complex dynamical networks studied in this paper is composed of many nodes coupled with each other, where each node is a dynamic basic unit with detailed content, which is modeled by the vector differential equation. Next, based on the Lyapunov stability theory, an appropriate adaptive control scheme and the adaptive coupling are designed for the controller plant, so that the controlled plant can asymptotically track its model following target. Furthermore, each node has a different model following target, which is actually achieve multi-target tracking control. Finally, numerical simulation is given to verify the effectiveness and correctness of the control scheme proposed in this paper.","PeriodicalId":441511,"journal":{"name":"2022 4th International Conference on Control and Robotics (ICCR)","volume":"212 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":"122454726","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/ICCR55715.2022.10053914
Puyang Liu, Song Xiao, Jie Liu, Chuanming Sun, Zuoqin Zhang, Junzhang Duan, Ye Cao, Jie Yu
The insulation section is an important part of the track circuit and plays a great role in judging the position of the train and the electrical isolation of the track circuit signal. As high-speed railway continue to increase speed, maintaining the high-speed operation of the train will inevitably require a larger traction current. When the train passes through the insulation section, the large rail current will frequently appear to be briefly disconnected, and the arcing phenomenon will often occur between the wheel and rail. Due to the high traction rail current and overvoltage caused by the higher grade arc will burn the insulation section, resulting in damage to the insulation section, which will greatly affect the operation safety of the train when it is serious. Based on the “train-rail” circuit model constructed, this paper discusses that when the running speed of the train is 100km/h, the switch and resistance are connected in series between adjacent rails. Then the rail current when the train passes through the insulation section is reduced by controlling the resistance, so as to reduce the level of arcing, reduce the impact of current on the rail, and effectively ensure the safe and stable operation of the train.
{"title":"Rail Current Suppression Strategy for Trains Passing through Insulation Section","authors":"Puyang Liu, Song Xiao, Jie Liu, Chuanming Sun, Zuoqin Zhang, Junzhang Duan, Ye Cao, Jie Yu","doi":"10.1109/ICCR55715.2022.10053914","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053914","url":null,"abstract":"The insulation section is an important part of the track circuit and plays a great role in judging the position of the train and the electrical isolation of the track circuit signal. As high-speed railway continue to increase speed, maintaining the high-speed operation of the train will inevitably require a larger traction current. When the train passes through the insulation section, the large rail current will frequently appear to be briefly disconnected, and the arcing phenomenon will often occur between the wheel and rail. Due to the high traction rail current and overvoltage caused by the higher grade arc will burn the insulation section, resulting in damage to the insulation section, which will greatly affect the operation safety of the train when it is serious. Based on the “train-rail” circuit model constructed, this paper discusses that when the running speed of the train is 100km/h, the switch and resistance are connected in series between adjacent rails. Then the rail current when the train passes through the insulation section is reduced by controlling the resistance, so as to reduce the level of arcing, reduce the impact of current on the rail, and effectively ensure the safe and stable operation of the train.","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":"115547930","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/ICCR55715.2022.10053878
Yuge Xu, Shuqiao Yang, Xie Zhang, Ziyi Xie
Steel surface defects Detection is crucial to improving the quality of steel production. However, the high-speed production lines, defect diversification, and tiny defects make the detection of steel surface defects difficult. This paper presents a steel surface defects detection model based on an improved Faster R-CNN. Firstly, to improve the generalization of the model, the ResNet50 network is replaced by the RegNet network. Then the transformer spatial attention is utilized to make the network focus more on the targets. Finally, transfer learning, multi-scale training, and cosine annealing learning rate are used to further improve the detection accuracy. Compared with the other nine models, the proposed model has superior performance in the simulation results. The improved model can effectively improve the accuracy of steel surface defects detection.
{"title":"Steel Surface Defects Detection Based on Improved Faster R-CNN","authors":"Yuge Xu, Shuqiao Yang, Xie Zhang, Ziyi Xie","doi":"10.1109/ICCR55715.2022.10053878","DOIUrl":"https://doi.org/10.1109/ICCR55715.2022.10053878","url":null,"abstract":"Steel surface defects Detection is crucial to improving the quality of steel production. However, the high-speed production lines, defect diversification, and tiny defects make the detection of steel surface defects difficult. This paper presents a steel surface defects detection model based on an improved Faster R-CNN. Firstly, to improve the generalization of the model, the ResNet50 network is replaced by the RegNet network. Then the transformer spatial attention is utilized to make the network focus more on the targets. Finally, transfer learning, multi-scale training, and cosine annealing learning rate are used to further improve the detection accuracy. Compared with the other nine models, the proposed model has superior performance in the simulation results. The improved model can effectively improve the accuracy of steel surface defects detection.","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":"129593698","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：