Pub Date : 2023-06-01DOI: 10.53106/199115992023063403003
Chen-Wei Feng Chen-Wei Feng, Xian-Guo Lu Chen-Wei Feng, Yu Sun Xian-Guo Lu, Huang-Bin Zeng Yu Sun, Zhuo Li Huang-Bin Zeng
As a special Mobile Ad-hoc Network (MANET), Vehicular Ad-hoc Network (VANET) plays a very important role in the future intelligent transportation system. In order to solve the problems of unstable communication connection, fast network topology change and low communication resource utilization caused by high vehicle mobility in VANET, a low-complexity resource allocation algorithm based on vehicle cluster is proposed. Firstly, considering the speed, position and moving direction of the vehicles, a vehicle clustering algorithm based on movement consistency is proposed to cluster the vehicles and keep the vehicle cluster stable. Secondly, a low-complexity resource allocation algorithm is proposed to improve the utilization rate of communication resources, which is constrained by the interference caused by the vehicle clusters to the cellular users. Simulation results show that the proposed algorithm has low complexity and can better maintain the stability of vehicle clusters and improve the system capacity in the common complex Internet of Vehicles (IoV) scenarios in cities.
{"title":"Vehicle Clustering and Resource Allocation Algorithm Based on Cellular Network","authors":"Chen-Wei Feng Chen-Wei Feng, Xian-Guo Lu Chen-Wei Feng, Yu Sun Xian-Guo Lu, Huang-Bin Zeng Yu Sun, Zhuo Li Huang-Bin Zeng","doi":"10.53106/199115992023063403003","DOIUrl":"https://doi.org/10.53106/199115992023063403003","url":null,"abstract":"\u0000 As a special Mobile Ad-hoc Network (MANET), Vehicular Ad-hoc Network (VANET) plays a very important role in the future intelligent transportation system. In order to solve the problems of unstable communication connection, fast network topology change and low communication resource utilization caused by high vehicle mobility in VANET, a low-complexity resource allocation algorithm based on vehicle cluster is proposed. Firstly, considering the speed, position and moving direction of the vehicles, a vehicle clustering algorithm based on movement consistency is proposed to cluster the vehicles and keep the vehicle cluster stable. Secondly, a low-complexity resource allocation algorithm is proposed to improve the utilization rate of communication resources, which is constrained by the interference caused by the vehicle clusters to the cellular users. Simulation results show that the proposed algorithm has low complexity and can better maintain the stability of vehicle clusters and improve the system capacity in the common complex Internet of Vehicles (IoV) scenarios in cities.\u0000 \u0000","PeriodicalId":345067,"journal":{"name":"電腦學刊","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116817702","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 : 2023-06-01DOI: 10.53106/199115992023063403009
Yuanjiang Hu Yuanjiang Hu, Aisen Yang Yuanjiang Hu, Zonghong Zhang Aisen Yang, Na Qin Zonghong Zhang
With the development of high-speed trains in recent years, security issues have received more attention. Automatic visual inspection of the train operation system for detecting abnormalities has become a fundamental element to guarantee the safety of the train operation. Train body sign patterns like the loss and fracture of signs and lock catch (SLC) on the electrical box cover (EBC) affect the regular operation of the train electrical system. In this paper, to ensure the safe operation of the train, a novel method combining a faster region-based convolutional neural network (Faster R-CNN) and similarity metrics is proposed to detect the abnormality of SLCs on train EBC. First, the positions of body train signs of multiple sizes are located by Faster R-CNN. Then, the regions of interest (ROI) are cut out and resized to the same size as the corresponding template images. Finally, by similarity measures, the status of the train body sign pattern is judged by comparing with the given threshold similarity value between ROIs and the template images. It is worth noting that the combination of Faster R-CNN and cosine similarity renders high accuracy in small target detection and strong robustness in image similarity comparison. The effectiveness of the proposed fault detection method and its superiority over the other types of combined methods are verified by actual experiments on the train of Guangzhou Metro Line 2.
{"title":"Fault Diagnosis of Train Body Sign Abnormal Pattern with Deep Learning Based Target Detection","authors":"Yuanjiang Hu Yuanjiang Hu, Aisen Yang Yuanjiang Hu, Zonghong Zhang Aisen Yang, Na Qin Zonghong Zhang","doi":"10.53106/199115992023063403009","DOIUrl":"https://doi.org/10.53106/199115992023063403009","url":null,"abstract":"\u0000 With the development of high-speed trains in recent years, security issues have received more attention. Automatic visual inspection of the train operation system for detecting abnormalities has become a fundamental element to guarantee the safety of the train operation. Train body sign patterns like the loss and fracture of signs and lock catch (SLC) on the electrical box cover (EBC) affect the regular operation of the train electrical system. In this paper, to ensure the safe operation of the train, a novel method combining a faster region-based convolutional neural network (Faster R-CNN) and similarity metrics is proposed to detect the abnormality of SLCs on train EBC. First, the positions of body train signs of multiple sizes are located by Faster R-CNN. Then, the regions of interest (ROI) are cut out and resized to the same size as the corresponding template images. Finally, by similarity measures, the status of the train body sign pattern is judged by comparing with the given threshold similarity value between ROIs and the template images. It is worth noting that the combination of Faster R-CNN and cosine similarity renders high accuracy in small target detection and strong robustness in image similarity comparison. The effectiveness of the proposed fault detection method and its superiority over the other types of combined methods are verified by actual experiments on the train of Guangzhou Metro Line 2.\u0000 \u0000","PeriodicalId":345067,"journal":{"name":"電腦學刊","volume":"17 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114517131","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 : 2023-06-01DOI: 10.53106/199115992023063403016
Xu-Nan Tan Xu-Nan Tan
Human activity recognition (HAR) based on wearable devices is an emerging field of great interest. HAR can provide additional information on a human subject’s physical status. Utilising new technologies for HAR will become very meaningful with the development of deep learning. This study aims to mine deep learning models for HAR prediction with the highest accuracy on the basis of time-series data collected by mobile wearable devices. To this end, convolutional neural networks (CNN) and long short-term memory neural networks (LSTM) are combined in a deep network model to extract behavioural facts. The proposed CNN model contains two convolutional layers and a maximum pooling layer, and batch normalisation is added after each convolutional layer to improve convergence speed and avoid overfitting. This structure yields significant results in terms of performance. The model is evaluated on the MHEALTH dataset with a test set accuracy of 99.61% and can be used for the intelligent recognition of human activity. The results of this study show that the proposed model has better robustness and motion pattern detection capability compared to other models.
{"title":"Human Activity Recognition Based on CNN and LSTM","authors":"Xu-Nan Tan Xu-Nan Tan","doi":"10.53106/199115992023063403016","DOIUrl":"https://doi.org/10.53106/199115992023063403016","url":null,"abstract":"\u0000 Human activity recognition (HAR) based on wearable devices is an emerging field of great interest. HAR can provide additional information on a human subject’s physical status. Utilising new technologies for HAR will become very meaningful with the development of deep learning. This study aims to mine deep learning models for HAR prediction with the highest accuracy on the basis of time-series data collected by mobile wearable devices. To this end, convolutional neural networks (CNN) and long short-term memory neural networks (LSTM) are combined in a deep network model to extract behavioural facts. The proposed CNN model contains two convolutional layers and a maximum pooling layer, and batch normalisation is added after each convolutional layer to improve convergence speed and avoid overfitting. This structure yields significant results in terms of performance. The model is evaluated on the MHEALTH dataset with a test set accuracy of 99.61% and can be used for the intelligent recognition of human activity. The results of this study show that the proposed model has better robustness and motion pattern detection capability compared to other models.\u0000 \u0000","PeriodicalId":345067,"journal":{"name":"電腦學刊","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116372359","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}
Ceramics have gradually occupied a more significant proportion in the art market and daily life in recent years. Therefore, the identification and anti-counterfeiting of ceramics have become more important with the continuous improvement of counterfeit ceramics. However, it is difficult for traditional ceramic identification and anti-counterfeiting technology to make instant, accurate and efficient identifications. Hence, based on the speed-ed up robust feature (SURF) algorithm, this paper proposes to take the microscopic surface features of ceramic images as the unique identifier for ceramic. In addition, blockchain was combined with distributed storage to ensure the security and reliability of these micro-characteristic data. At any time, ceramic images to be identified can be compared and verified with these images stored on the blockchain, and hence to determine the authenticity of the ceramics. Experimental results show that the proposed method has a high recognition rate and good robustness to problems. Compared with the traditional feature extraction methods, the efficiency and accuracy of proposed algorithm have been improved. The matching similarity rate between most imitations and genuine products using the proposed algorithm will not exceed 15%, thus accurately identifying imitations to achieve the anti-counterfeiting of ceramics.
{"title":"A Recognition Method of Ceramic Microcosmic Images Based on SURF and Blockchain","authors":"You-Dong Wang You-Dong Wang, Xing Xu You-Dong Wang, Xi-En Cheng Xing Xu","doi":"10.53106/199115992023063403011","DOIUrl":"https://doi.org/10.53106/199115992023063403011","url":null,"abstract":"\u0000 Ceramics have gradually occupied a more significant proportion in the art market and daily life in recent years. Therefore, the identification and anti-counterfeiting of ceramics have become more important with the continuous improvement of counterfeit ceramics. However, it is difficult for traditional ceramic identification and anti-counterfeiting technology to make instant, accurate and efficient identifications. Hence, based on the speed-ed up robust feature (SURF) algorithm, this paper proposes to take the microscopic surface features of ceramic images as the unique identifier for ceramic. In addition, blockchain was combined with distributed storage to ensure the security and reliability of these micro-characteristic data. At any time, ceramic images to be identified can be compared and verified with these images stored on the blockchain, and hence to determine the authenticity of the ceramics. Experimental results show that the proposed method has a high recognition rate and good robustness to problems. Compared with the traditional feature extraction methods, the efficiency and accuracy of proposed algorithm have been improved. The matching similarity rate between most imitations and genuine products using the proposed algorithm will not exceed 15%, thus accurately identifying imitations to achieve the anti-counterfeiting of ceramics.\u0000 \u0000","PeriodicalId":345067,"journal":{"name":"電腦學刊","volume":"51 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116434683","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 : 2023-06-01DOI: 10.53106/199115992023063403001
Wei Li Wei Li, Yang Gao Wei Li, Jun Chen Yang Gao, Si-Yi Niu Jun Chen, Jia-Hao Jiang Si-Yi Niu, Qi Li Jia-Hao Jiang
In this paper, we propose a time sequential IC3D convolutional neural network approach for hand gesture recognition based on frequency modulated continuous wave (FMCW) radar. Firstly, the FMCW radar is used to collect the echoes of human hand gestures. A two-dimensional fast Fourier transform calculates the range and velocity information of hand gestures in each frame signal to construct the Range-Doppler heat map dataset of hand gestures. Then, we design an IC3D network for feature extraction and classification of the dynamic gesture heat map. Finally, the experiment results show that the gesture recognition system designed in this paper effectively solves the problems of the difficulty of human gesture feature extraction and low utilization of time series information, and the average recognition accuracy rate can reach more than 99.8%.
{"title":"Human Gesture Recognition Based on Millimeter-Wave Radar Using Improved C3D Convolutional Neural Network","authors":"Wei Li Wei Li, Yang Gao Wei Li, Jun Chen Yang Gao, Si-Yi Niu Jun Chen, Jia-Hao Jiang Si-Yi Niu, Qi Li Jia-Hao Jiang","doi":"10.53106/199115992023063403001","DOIUrl":"https://doi.org/10.53106/199115992023063403001","url":null,"abstract":"\u0000 In this paper, we propose a time sequential IC3D convolutional neural network approach for hand gesture recognition based on frequency modulated continuous wave (FMCW) radar. Firstly, the FMCW radar is used to collect the echoes of human hand gestures. A two-dimensional fast Fourier transform calculates the range and velocity information of hand gestures in each frame signal to construct the Range-Doppler heat map dataset of hand gestures. Then, we design an IC3D network for feature extraction and classification of the dynamic gesture heat map. Finally, the experiment results show that the gesture recognition system designed in this paper effectively solves the problems of the difficulty of human gesture feature extraction and low utilization of time series information, and the average recognition accuracy rate can reach more than 99.8%.\u0000 \u0000","PeriodicalId":345067,"journal":{"name":"電腦學刊","volume":"584 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123040223","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 : 2023-06-01DOI: 10.53106/199115992023063403024
Han Gao Han Gao, Dan Wang Han Gao, Ying He Dan Wang, Yang-Yang Yu Ying He, Bai-Jun Gao Yang-Yang Yu
Analog circuit faults are the main cause of performance degradation or paralysis in integrated circuit systems. However, due to the complex causes and diverse manifestations of circuit faults themselves, traditional methods have high difficulty in identifying typical faults in analog circuits and low recognition accuracy. This article constructs an improved ResNet deep feature recognition network model and establishes one-dimensional and two-dimensional fault information sources. Finally, particle swarm optimization algorithm is used to search for the optimal parameters solved by the model, ultimately achieving improvements in the accuracy and recognition speed of analog circuit fault diagnosis. Finally, through experimental verification, the recognition accuracy of typical fault C2 reached 99.6%, proving the effectiveness of the method proposed in this paper.
{"title":"Strategy for Identifying Analog Circuit Faults Using Improved Neural Network Algorithms","authors":"Han Gao Han Gao, Dan Wang Han Gao, Ying He Dan Wang, Yang-Yang Yu Ying He, Bai-Jun Gao Yang-Yang Yu","doi":"10.53106/199115992023063403024","DOIUrl":"https://doi.org/10.53106/199115992023063403024","url":null,"abstract":"\u0000 Analog circuit faults are the main cause of performance degradation or paralysis in integrated circuit systems. However, due to the complex causes and diverse manifestations of circuit faults themselves, traditional methods have high difficulty in identifying typical faults in analog circuits and low recognition accuracy. This article constructs an improved ResNet deep feature recognition network model and establishes one-dimensional and two-dimensional fault information sources. Finally, particle swarm optimization algorithm is used to search for the optimal parameters solved by the model, ultimately achieving improvements in the accuracy and recognition speed of analog circuit fault diagnosis. Finally, through experimental verification, the recognition accuracy of typical fault C2 reached 99.6%, proving the effectiveness of the method proposed in this paper. \u0000","PeriodicalId":345067,"journal":{"name":"電腦學刊","volume":"4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123885125","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 : 2023-06-01DOI: 10.53106/199115992023063403021
Jin-Ping Du Jin-Ping Du, Xiao-Fei Wu Jin-Ping Du, Jian Wang Xiao-Fei Wu, Dong-Liang Fan Jian Wang, Qian-Han Zhang Dong-Liang Fan
This article proposes a multi-objective function that includes AGV running time, production workshop energy consumption, and machine running efficiency, in response to the problems of path conflicts, single planning objectives, and isolation of planning stages in the current flexible production workshop AGV car planning. Then, the flying mouse algorithm is used to solve the problem using multiple functions. In order to avoid falling into local optima during the solving process, a simulated annealing strategy is incorporated into the flying mouse algorithm. Finally, taking the production of new energy vehicle on-board batteries as an example, a collaborative planning analysis was conducted using the method proposed in this paper. The results showed that the algorithm proposed in this paper can save 30% of running time and improve machine operating efficiency by 22.7%.
{"title":"Collaborative Planning Method for Flexible Production Workshop Equipment and AGV Trolley Based on Artificial Intelligence Algorithms","authors":"Jin-Ping Du Jin-Ping Du, Xiao-Fei Wu Jin-Ping Du, Jian Wang Xiao-Fei Wu, Dong-Liang Fan Jian Wang, Qian-Han Zhang Dong-Liang Fan","doi":"10.53106/199115992023063403021","DOIUrl":"https://doi.org/10.53106/199115992023063403021","url":null,"abstract":"\u0000 This article proposes a multi-objective function that includes AGV running time, production workshop energy consumption, and machine running efficiency, in response to the problems of path conflicts, single planning objectives, and isolation of planning stages in the current flexible production workshop AGV car planning. Then, the flying mouse algorithm is used to solve the problem using multiple functions. In order to avoid falling into local optima during the solving process, a simulated annealing strategy is incorporated into the flying mouse algorithm. Finally, taking the production of new energy vehicle on-board batteries as an example, a collaborative planning analysis was conducted using the method proposed in this paper. The results showed that the algorithm proposed in this paper can save 30% of running time and improve machine operating efficiency by 22.7%. \u0000","PeriodicalId":345067,"journal":{"name":"電腦學刊","volume":"98 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116182810","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 : 2023-06-01DOI: 10.53106/199115992023063403013
Chengcheng Zou Chengcheng Zou, Damin Zhang Chengcheng Zou, Linna Zhang Damin Zhang
To address the limited and time delay disaster communication, a joint optimization scheme integrates the advantages of differential evolution algorithm (DE) and naked mole- rat algorithm (NMR), and proposes Lévy and sigmoidal DE-NMR, namely LS-DN. LS-DN applies the Lévy flight parameters of adaptive features and sigmoidal selection factor (λ) to the worker of NMR phase, and optimizes the crossover rate (CR) and variation parameter (F) in the DE algorithm, to obtain a balance the exploration and development capabilities. The proposed LS-DN algorithm is used to optimize the user aggregation scheme, since an effect aggregation of disaster victims can reduce power consumption and improve system performance. An value of power external function (Cfn ) is defined for each disaster victim, which is expressed as the system power consumption value for each disaster victim under different aggregation schemes. To minimize the microcell power without deteriorating the quality of service (QoS), it is demonstrated by analyzing the relevant characteristics of non-orthogonal multiple access（NOMA）disaster communication that the power consumption strongly depends on user aggregation method and power allocation. The significance of joint optimization for improving the performance of NOMA disaster communication systems is also emphasized. Simulation results show that LS-DN is able to significantly reduce the power consumption of the system. With the application of LS-DN, the throughput of NOMA system increases by 65% compared to the conventional orthogonal multiple access (OMA) system.
{"title":"LS-DN Algorithm Based User Matching and Power Minimization in NOMA Disaster Communication","authors":"Chengcheng Zou Chengcheng Zou, Damin Zhang Chengcheng Zou, Linna Zhang Damin Zhang","doi":"10.53106/199115992023063403013","DOIUrl":"https://doi.org/10.53106/199115992023063403013","url":null,"abstract":"\u0000 To address the limited and time delay disaster communication, a joint optimization scheme integrates the advantages of differential evolution algorithm (DE) and naked mole- rat algorithm (NMR), and proposes Lévy and sigmoidal DE-NMR, namely LS-DN. LS-DN applies the Lévy flight parameters of adaptive features and sigmoidal selection factor (λ) to the worker of NMR phase, and optimizes the crossover rate (CR) and variation parameter (F) in the DE algorithm, to obtain a balance the exploration and development capabilities. The proposed LS-DN algorithm is used to optimize the user aggregation scheme, since an effect aggregation of disaster victims can reduce power consumption and improve system performance. An value of power external function (Cfn ) is defined for each disaster victim, which is expressed as the system power consumption value for each disaster victim under different aggregation schemes. To minimize the microcell power without deteriorating the quality of service (QoS), it is demonstrated by analyzing the relevant characteristics of non-orthogonal multiple access（NOMA）disaster communication that the power consumption strongly depends on user aggregation method and power allocation. The significance of joint optimization for improving the performance of NOMA disaster communication systems is also emphasized. Simulation results show that LS-DN is able to significantly reduce the power consumption of the system. With the application of LS-DN, the throughput of NOMA system increases by 65% compared to the conventional orthogonal multiple access (OMA) system.\u0000 \u0000","PeriodicalId":345067,"journal":{"name":"電腦學刊","volume":"22 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125951312","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 : 2023-06-01DOI: 10.53106/199115992023063403004
Qiang Yuan Qiang Yuan, Shuai-Shuai Liu Qiang Yuan, Bang-Yu Wang Shuai-Shuai Liu, Dang-Wei Han Bang-Yu Wang, Sai-Nan Du Dang-Wei Han, Da-Xu Zhao Sai-Nan Du
In this paper, we propose an algorithm model PearlNet and the corresponding detection dataset for freshwater pearls detection, to increase the Degree of Automation and improve the efficiency of existing detection methods based on pearl colors and shapes. PearlNet based on CenterNet. According to the characteristics of the small target of freshwater pearls, the minimum size module of the network is deleted, and the attention mechanism is added at the same time, ignoring the irrelevant background information and focusing on the pearl feature information, which improves the accuracy of recognition. In the transport convolution process, the image quality effect caused by upsampling is reduced by data fusion. The experimental results proved that the PearlNet has a recognition accuracy of 98.4%, which is 15.43%, 9.05% and 5.2% higher than that of CenterNet, Yolo V3 and SSD. PearlNet can accurately identify the color and shape of pearls, which provides a reference for freshwater pearl identification and detection.
{"title":"Pearl Detection Based on PearlNet","authors":"Qiang Yuan Qiang Yuan, Shuai-Shuai Liu Qiang Yuan, Bang-Yu Wang Shuai-Shuai Liu, Dang-Wei Han Bang-Yu Wang, Sai-Nan Du Dang-Wei Han, Da-Xu Zhao Sai-Nan Du","doi":"10.53106/199115992023063403004","DOIUrl":"https://doi.org/10.53106/199115992023063403004","url":null,"abstract":"\u0000 In this paper, we propose an algorithm model PearlNet and the corresponding detection dataset for freshwater pearls detection, to increase the Degree of Automation and improve the efficiency of existing detection methods based on pearl colors and shapes. PearlNet based on CenterNet. According to the characteristics of the small target of freshwater pearls, the minimum size module of the network is deleted, and the attention mechanism is added at the same time, ignoring the irrelevant background information and focusing on the pearl feature information, which improves the accuracy of recognition. In the transport convolution process, the image quality effect caused by upsampling is reduced by data fusion. The experimental results proved that the PearlNet has a recognition accuracy of 98.4%, which is 15.43%, 9.05% and 5.2% higher than that of CenterNet, Yolo V3 and SSD. PearlNet can accurately identify the color and shape of pearls, which provides a reference for freshwater pearl identification and detection.\u0000 \u0000","PeriodicalId":345067,"journal":{"name":"電腦學刊","volume":"38 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"122549447","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 : 2023-06-01DOI: 10.53106/199115992023063403006
Li-Hong Deng Li-Hong Deng, Fei Deng Li-Hong Deng, Ge-Xiang Chiou Fei Deng, Qiang Yang Ge-Xiang Chiou
The feature extraction ability of lightweight convolutional neural networks in speaker recognition systems is weak. And recognition accuracy is poor. Many methods use deeper, wider, and more complex network structures to improve the feature extraction ability. But it makes the parameters and inference time increase exponentially. In the paper, we introduce Res2Net in target detection task to speaker recognition task and verify its effectiveness and robustness in the speaker recognition task. And we improved and proposed FullRes2Net. It has better multi-scale feature extraction ability without increasing the number of parameters. Then, we proposed the mixed time-frequency channel attention to solve the problems of existing attention methods to improve the shortcomings of convolution itself and further enhance the feature extraction ability of convolutional neural networks. Experiments were conducted on the Voxceleb dataset. The results show that the MTFC-FullRes2Net end-to-end speaker recognition system proposed in this paper effectively improves the feature extraction and generalization ability of the Res2Net. Compared to Res2Net, MTFC-FullRes2Net performance improves by 31.5%. And Compared to ThinResNet-50, RawNet, CNN+Transformer and Y-vector, MTFC-FullRes2Net performance is improved by 56.5%, 14.1%, 16.7% and 23.4%, respectively. And it is superior to state-of-the-art speaker recognition systems that use complex structures. It is a lightweight and more efficient end-to-end architecture and is also more suitable for practical application.
{"title":"End-to-end Speaker Recognition Based on MTFC-FullRes2Net","authors":"Li-Hong Deng Li-Hong Deng, Fei Deng Li-Hong Deng, Ge-Xiang Chiou Fei Deng, Qiang Yang Ge-Xiang Chiou","doi":"10.53106/199115992023063403006","DOIUrl":"https://doi.org/10.53106/199115992023063403006","url":null,"abstract":"\u0000 The feature extraction ability of lightweight convolutional neural networks in speaker recognition systems is weak. And recognition accuracy is poor. Many methods use deeper, wider, and more complex network structures to improve the feature extraction ability. But it makes the parameters and inference time increase exponentially. In the paper, we introduce Res2Net in target detection task to speaker recognition task and verify its effectiveness and robustness in the speaker recognition task. And we improved and proposed FullRes2Net. It has better multi-scale feature extraction ability without increasing the number of parameters. Then, we proposed the mixed time-frequency channel attention to solve the problems of existing attention methods to improve the shortcomings of convolution itself and further enhance the feature extraction ability of convolutional neural networks. Experiments were conducted on the Voxceleb dataset. The results show that the MTFC-FullRes2Net end-to-end speaker recognition system proposed in this paper effectively improves the feature extraction and generalization ability of the Res2Net. Compared to Res2Net, MTFC-FullRes2Net performance improves by 31.5%. And Compared to ThinResNet-50, RawNet, CNN+Transformer and Y-vector, MTFC-FullRes2Net performance is improved by 56.5%, 14.1%, 16.7% and 23.4%, respectively. And it is superior to state-of-the-art speaker recognition systems that use complex structures. It is a lightweight and more efficient end-to-end architecture and is also more suitable for practical application.\u0000 \u0000","PeriodicalId":345067,"journal":{"name":"電腦學刊","volume":"143 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-06-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128121157","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：