This paper presents a design method of general modular digital target simulation system, and realizes the software based on this method. This method includes target vulnerability model import sub module and target vulnerability model construction sub module. It mainly completes the import or construction of target vulnerability model, target location determination, shape feature identification, identification of key components and non-key components, division of killing mode of key components, establishment of geometric model of components, etc. It can provide target vulnerability related data for damage effectiveness evaluation and fire plan.
{"title":"Design Method of Universal Modular Digital Target Simulation System","authors":"Yong Sun, W. Yang, Guangzhao Lu, Jinqing Zhao, Xiaoyue Wang, Ji-rong Xue","doi":"10.1109/cvidliccea56201.2022.9825333","DOIUrl":"https://doi.org/10.1109/cvidliccea56201.2022.9825333","url":null,"abstract":"This paper presents a design method of general modular digital target simulation system, and realizes the software based on this method. This method includes target vulnerability model import sub module and target vulnerability model construction sub module. It mainly completes the import or construction of target vulnerability model, target location determination, shape feature identification, identification of key components and non-key components, division of killing mode of key components, establishment of geometric model of components, etc. It can provide target vulnerability related data for damage effectiveness evaluation and fire plan.","PeriodicalId":23649,"journal":{"name":"Vision","volume":"192 1","pages":"931-934"},"PeriodicalIF":0.0,"publicationDate":"2022-05-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"89085861","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 : 2022-05-20DOI: 10.1109/cvidliccea56201.2022.9825237
Haixin Lin, Hongzhi Ma, Weibin Gong, Chao Wang
Frontal face image recognition is the main target of traditional face recognition.The deflection of the human face often causes the dislocation of the facial features,which leads to the reduction of the recognition accuracy of the non-frontal face.To solve the above problems,a non-frontal face recognition model based on generative adversarial network is proposed.In this model,the angle information is encoded separately by using a two-channel generator and auto-coding network,and the non-frontal face image in natural environment is corrected to obtain the frontal face image.Through the multi-discriminator mechanism of facial attention,we set discriminators in the eye, eyebrow, nose, mouth and the whole area of the face image so as to retain the details of the face to the maximum extent while ensuring the clarity of image.Then the corrected face features are extracted by Facenet and MTCNN to obtain the non-frontal face recognition results.The model is validated on multi-PIE dataset and CFP dataset.The results show that the accuracy of non-frontal face recognition is improved by 1% in CFP dataset compared with VGG-FACE, TP- CNN and HPN.
{"title":"Non-frontal face recognition method with a side-face-correction generative adversarial networks","authors":"Haixin Lin, Hongzhi Ma, Weibin Gong, Chao Wang","doi":"10.1109/cvidliccea56201.2022.9825237","DOIUrl":"https://doi.org/10.1109/cvidliccea56201.2022.9825237","url":null,"abstract":"Frontal face image recognition is the main target of traditional face recognition.The deflection of the human face often causes the dislocation of the facial features,which leads to the reduction of the recognition accuracy of the non-frontal face.To solve the above problems,a non-frontal face recognition model based on generative adversarial network is proposed.In this model,the angle information is encoded separately by using a two-channel generator and auto-coding network,and the non-frontal face image in natural environment is corrected to obtain the frontal face image.Through the multi-discriminator mechanism of facial attention,we set discriminators in the eye, eyebrow, nose, mouth and the whole area of the face image so as to retain the details of the face to the maximum extent while ensuring the clarity of image.Then the corrected face features are extracted by Facenet and MTCNN to obtain the non-frontal face recognition results.The model is validated on multi-PIE dataset and CFP dataset.The results show that the accuracy of non-frontal face recognition is improved by 1% in CFP dataset compared with VGG-FACE, TP- CNN and HPN.","PeriodicalId":23649,"journal":{"name":"Vision","volume":"625 1","pages":"563-567"},"PeriodicalIF":0.0,"publicationDate":"2022-05-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"78971103","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 : 2022-05-20DOI: 10.1109/cvidliccea56201.2022.9824224
Keun-Jong Lyu, Haizhang Yan
This paper uses the machine vision method to identify the skirt module. We have constructed three kinds of machine recognition models of skirt profile processing, structure analysis of style drawing, and size estimation. The author constructs a relatively complete image recognition system for dress pattern drawing. In addition, we conducted an effect evaluation with a certain number of samples at the later stage of the experiment. This study has A good effect in distinguishing an A-type skirt from an H-type skirt, identifying the reasonable degree and length of the skirt, and determining the quantity statistics of each component element in the skirt pattern diagram.
{"title":"Identification Method of Dress Pattern Drawing based on Machine Vision Algorithm","authors":"Keun-Jong Lyu, Haizhang Yan","doi":"10.1109/cvidliccea56201.2022.9824224","DOIUrl":"https://doi.org/10.1109/cvidliccea56201.2022.9824224","url":null,"abstract":"This paper uses the machine vision method to identify the skirt module. We have constructed three kinds of machine recognition models of skirt profile processing, structure analysis of style drawing, and size estimation. The author constructs a relatively complete image recognition system for dress pattern drawing. In addition, we conducted an effect evaluation with a certain number of samples at the later stage of the experiment. This study has A good effect in distinguishing an A-type skirt from an H-type skirt, identifying the reasonable degree and length of the skirt, and determining the quantity statistics of each component element in the skirt pattern diagram.","PeriodicalId":23649,"journal":{"name":"Vision","volume":"32 1","pages":"76-79"},"PeriodicalIF":0.0,"publicationDate":"2022-05-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"76514713","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 : 2022-05-20DOI: 10.1109/cvidliccea56201.2022.9824088
Yujie Lu, Yidi Lu
Collaborative filtering is widely applied in recommendation systems. The traditional method usually adopts the cosine similarity algorithm or Pearson algorithm, but a sparse rating matrix may lead to inaccurate recommendation results. The optimized algorithm adds penalty terms according to the number of score vector elements to reduce the impact of sparsity. More purchase behaviors are taken into account in the optimization algorithm, including user activity, product popularity, and the time cost of user preferences. Due to the validity of the data set, the top-k method is adopted to select k users with the highest similarity (1) as the recommendation basis. Compared with the traditional method, the numerical results have a lower root mean squared error, and the algorithm execution time is significantly shortened. The optimized collaborative filtering algorithm can effectively alleviate the impact of sparsity and consider more purchasing behaviors, thus improving the algorithm efficiency and rating reliability of the book recommendation system.
{"title":"Book recommendation system based on an optimized collaborative filtering algorithm","authors":"Yujie Lu, Yidi Lu","doi":"10.1109/cvidliccea56201.2022.9824088","DOIUrl":"https://doi.org/10.1109/cvidliccea56201.2022.9824088","url":null,"abstract":"Collaborative filtering is widely applied in recommendation systems. The traditional method usually adopts the cosine similarity algorithm or Pearson algorithm, but a sparse rating matrix may lead to inaccurate recommendation results. The optimized algorithm adds penalty terms according to the number of score vector elements to reduce the impact of sparsity. More purchase behaviors are taken into account in the optimization algorithm, including user activity, product popularity, and the time cost of user preferences. Due to the validity of the data set, the top-k method is adopted to select k users with the highest similarity (1) as the recommendation basis. Compared with the traditional method, the numerical results have a lower root mean squared error, and the algorithm execution time is significantly shortened. The optimized collaborative filtering algorithm can effectively alleviate the impact of sparsity and consider more purchasing behaviors, thus improving the algorithm efficiency and rating reliability of the book recommendation system.","PeriodicalId":23649,"journal":{"name":"Vision","volume":"74 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2022-05-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"74089388","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 : 2022-05-20DOI: 10.1109/cvidliccea56201.2022.9824472
Yan Ma, Ruoyu Fang
In response to the problems of the traditional new energy vehicle power battery traceability system such as centralized easy tampering, data cannot be shared and lack of effective management, this paper proposes a blockchain-based new energy vehicle power battery supply chain traceability system. Analyzed the business processes in the power battery supply chain of production, vehicle, sales, power exchange and recycling, designed the system architecture according to the different needs of regulators, consumers, enterprises and other subjects, established the traceability information chain and database, and proposed a smart contract applicable to the system. The system relies on the characteristics of blockchain decentralization and on-chain data that cannot be tampered with, which protects the privacy of users and improves the reliability of the system while meeting the traceability needs of power batteries.
{"title":"Blockchain-based power battery traceability system for new energy vehicles","authors":"Yan Ma, Ruoyu Fang","doi":"10.1109/cvidliccea56201.2022.9824472","DOIUrl":"https://doi.org/10.1109/cvidliccea56201.2022.9824472","url":null,"abstract":"In response to the problems of the traditional new energy vehicle power battery traceability system such as centralized easy tampering, data cannot be shared and lack of effective management, this paper proposes a blockchain-based new energy vehicle power battery supply chain traceability system. Analyzed the business processes in the power battery supply chain of production, vehicle, sales, power exchange and recycling, designed the system architecture according to the different needs of regulators, consumers, enterprises and other subjects, established the traceability information chain and database, and proposed a smart contract applicable to the system. The system relies on the characteristics of blockchain decentralization and on-chain data that cannot be tampered with, which protects the privacy of users and improves the reliability of the system while meeting the traceability needs of power batteries.","PeriodicalId":23649,"journal":{"name":"Vision","volume":"16 1","pages":"248-251"},"PeriodicalIF":0.0,"publicationDate":"2022-05-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"88432935","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 : 2022-05-20DOI: 10.1109/cvidliccea56201.2022.9824964
Kengpeng Li, Fenfa Zhong, Lei Sun
Hyperspectral image denoising is an essential pre-processing task. In this paper, a multi-resolution gated network based on wavelet transform (WMRGNet) is proposed for removing mixed noise of hyperspectral images. Firstly, based on the fact that hyperspectral images have strong spectral correlation, a spatial-spectral information extraction module is designed to use the current noisy band and its adjacent bands as the input of WMRGNet. Secondly, aim to fully consider the spatial local and global information of hyperspectral images, a multi-resolution feature extraction module is proposed, applying the discrete wavelet transform to divide the resolution into four scales, and the residual blocks to extract information of different resolutions. In addition, a gated layer is introduced for cross-resolution information interaction to enhance the feature fusion. Finally, a high-resolution image reconstruction module with multiple residual blocks is employed to extract high-resolution features. In the simulated data set experiments, WMRGNet removes Gaussian, stripe and deadline noise and preserves the detailed information of the hyperspectral images.
{"title":"Hyperspectral Image Denoising Based on Multi-Resolution Gated Network with Wavelet Transform","authors":"Kengpeng Li, Fenfa Zhong, Lei Sun","doi":"10.1109/cvidliccea56201.2022.9824964","DOIUrl":"https://doi.org/10.1109/cvidliccea56201.2022.9824964","url":null,"abstract":"Hyperspectral image denoising is an essential pre-processing task. In this paper, a multi-resolution gated network based on wavelet transform (WMRGNet) is proposed for removing mixed noise of hyperspectral images. Firstly, based on the fact that hyperspectral images have strong spectral correlation, a spatial-spectral information extraction module is designed to use the current noisy band and its adjacent bands as the input of WMRGNet. Secondly, aim to fully consider the spatial local and global information of hyperspectral images, a multi-resolution feature extraction module is proposed, applying the discrete wavelet transform to divide the resolution into four scales, and the residual blocks to extract information of different resolutions. In addition, a gated layer is introduced for cross-resolution information interaction to enhance the feature fusion. Finally, a high-resolution image reconstruction module with multiple residual blocks is employed to extract high-resolution features. In the simulated data set experiments, WMRGNet removes Gaussian, stripe and deadline noise and preserves the detailed information of the hyperspectral images.","PeriodicalId":23649,"journal":{"name":"Vision","volume":"56 1","pages":"637-642"},"PeriodicalIF":0.0,"publicationDate":"2022-05-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"88935014","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 : 2022-05-20DOI: 10.1109/cvidliccea56201.2022.9824130
Lei Qiu, Yize Fan
smart noise jamming based on digital radio frequency memory (DRFM) has both suppression jamming and deception jamming effect, and is difficult to be effectively suppressed by traditional anti-jamming methods. In order to solve this problem, a smart noise jamming suppression method based on target position estimation is proposed. Firstly, the difference between tracking radar target returns and convolution smart noise jamming signals are analyzed. Then fractional Fourier transform (FRT) of linear frequency modulation (LFM) signal is derived, and the target position at the current frame is estimated based on the previous target position and velocity without jamming, followed by a filter in the FRET domain to suppress the jamming signal. Finally, the suppressed signal is transformed to time domain through inverse FRFT. Simulation results verified the feasibility of the proposed method with a high JSR.
{"title":"Smart Noise Jamming Suppression Method Based on Target Position Estimation","authors":"Lei Qiu, Yize Fan","doi":"10.1109/cvidliccea56201.2022.9824130","DOIUrl":"https://doi.org/10.1109/cvidliccea56201.2022.9824130","url":null,"abstract":"smart noise jamming based on digital radio frequency memory (DRFM) has both suppression jamming and deception jamming effect, and is difficult to be effectively suppressed by traditional anti-jamming methods. In order to solve this problem, a smart noise jamming suppression method based on target position estimation is proposed. Firstly, the difference between tracking radar target returns and convolution smart noise jamming signals are analyzed. Then fractional Fourier transform (FRT) of linear frequency modulation (LFM) signal is derived, and the target position at the current frame is estimated based on the previous target position and velocity without jamming, followed by a filter in the FRET domain to suppress the jamming signal. Finally, the suppressed signal is transformed to time domain through inverse FRFT. Simulation results verified the feasibility of the proposed method with a high JSR.","PeriodicalId":23649,"journal":{"name":"Vision","volume":"37 4 1","pages":"1-4"},"PeriodicalIF":0.0,"publicationDate":"2022-05-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87537313","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 : 2022-05-20DOI: 10.1109/cvidliccea56201.2022.9824137
Xiujun Li, LiMin Tang, Zhilin Zhang, Jinglong Wu, Qi Li
With the rapid development of the society, people’s cognitive ability is gradually declining under the pressure of study and work, those declining will bring a great impact on people’s daily life, work and study. Therefore, it has become a popular research field on the intervention of early declining cognitive ability. It has been proved that a brain-computer interface (BCI), as a new tool of neural feedback training, can improve the traditional neural feedback method’s efficiency. However, due to the weakness of EEG signals, the accuracy of character’s recognition in BCI system is still low, and the spelling paradigm is often dominated by the characters output, which makes it difficult for the subjects’ attention to continuously focus on the whole experiment process. In order to improve the recognition accuracy and effectively reduce subjects’ fatigue as well as increase the concentration of the participant, this study proposed a spelling paradigm that can more effectively stimulate the Event Related Potential (ERP) of the subject. In the end, a cognitive ability training system was designed for daily training. Through the system simulation and preliminary experiments, the average target recognition accuracy increased by 18.75% after the experiment, demonstrating the efficiency and effectiveness of the present cognitive ability training system.
随着社会的快速发展，人们的认知能力在学习和工作的压力下逐渐下降，这些下降会给人们的日常生活、工作和学习带来很大的影响。因此，早期认知能力衰退的干预已成为一个热门的研究领域。事实证明，脑机接口(BCI)作为一种新的神经反馈训练工具，可以提高传统神经反馈训练方法的效率。然而，由于脑电图信号的薄弱，BCI系统对字符识别的准确率仍然较低，拼写范式往往以字符输出为主，这使得被试的注意力难以持续集中在整个实验过程中。为了提高识别准确率，有效降低被试的疲劳程度，提高被试的注意力，本研究提出了一种能够更有效刺激被试事件相关电位(Event Related Potential, ERP)的拼写范式。最后，针对日常训练设计了认知能力训练系统。通过系统仿真和初步实验，实验后的平均目标识别准确率提高了18.75%，证明了本认知能力训练系统的效率和有效性。
{"title":"Attention and Memory Training System Based on Neural Feedback","authors":"Xiujun Li, LiMin Tang, Zhilin Zhang, Jinglong Wu, Qi Li","doi":"10.1109/cvidliccea56201.2022.9824137","DOIUrl":"https://doi.org/10.1109/cvidliccea56201.2022.9824137","url":null,"abstract":"With the rapid development of the society, people’s cognitive ability is gradually declining under the pressure of study and work, those declining will bring a great impact on people’s daily life, work and study. Therefore, it has become a popular research field on the intervention of early declining cognitive ability. It has been proved that a brain-computer interface (BCI), as a new tool of neural feedback training, can improve the traditional neural feedback method’s efficiency. However, due to the weakness of EEG signals, the accuracy of character’s recognition in BCI system is still low, and the spelling paradigm is often dominated by the characters output, which makes it difficult for the subjects’ attention to continuously focus on the whole experiment process. In order to improve the recognition accuracy and effectively reduce subjects’ fatigue as well as increase the concentration of the participant, this study proposed a spelling paradigm that can more effectively stimulate the Event Related Potential (ERP) of the subject. In the end, a cognitive ability training system was designed for daily training. Through the system simulation and preliminary experiments, the average target recognition accuracy increased by 18.75% after the experiment, demonstrating the efficiency and effectiveness of the present cognitive ability training system.","PeriodicalId":23649,"journal":{"name":"Vision","volume":"6 1","pages":"794-798"},"PeriodicalIF":0.0,"publicationDate":"2022-05-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"87541152","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 : 2022-05-20DOI: 10.1109/cvidliccea56201.2022.9824164
Peng Yang, Zhirong Peng, Wu-Long Wang
Aiming at a low-power embedded real-time sitting posture detection system, a real-time sitting posture detection system based on deep learning is designed. The system obtains the pressure of the sitting posture of the human body through a thin-film pressure sensor and the human body pressure in different sitting postures is collected and analyzed, and an analysis model is established under the Keras framework. Burn the model into STM32 through cubemax to realize real-time collection, analysis and detection of human sitting posture. Finally, the communication between the STM32 and the Android application is realized through the MQTT protocol, which realizes the real-time detection and discrimination of the sitting posture and gives the relevant sitting posture correction prompts.
{"title":"Sitting Posture Detection System Based on Keras Framework","authors":"Peng Yang, Zhirong Peng, Wu-Long Wang","doi":"10.1109/cvidliccea56201.2022.9824164","DOIUrl":"https://doi.org/10.1109/cvidliccea56201.2022.9824164","url":null,"abstract":"Aiming at a low-power embedded real-time sitting posture detection system, a real-time sitting posture detection system based on deep learning is designed. The system obtains the pressure of the sitting posture of the human body through a thin-film pressure sensor and the human body pressure in different sitting postures is collected and analyzed, and an analysis model is established under the Keras framework. Burn the model into STM32 through cubemax to realize real-time collection, analysis and detection of human sitting posture. Finally, the communication between the STM32 and the Android application is realized through the MQTT protocol, which realizes the real-time detection and discrimination of the sitting posture and gives the relevant sitting posture correction prompts.","PeriodicalId":23649,"journal":{"name":"Vision","volume":"36 1","pages":"170-174"},"PeriodicalIF":0.0,"publicationDate":"2022-05-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90368202","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 : 2022-05-20DOI: 10.1109/cvidliccea56201.2022.9824054
Xu Chu, Xiaoyang Liu
The detection of cell and microbes under the microscope is of great value in both clinical experiments and experimental teaching. However, the narrow field of view of conventional light microscopes and the problem of cell or microbial stacking make target detection a challenging task. In this paper, the YOLOv5 target detection method is improved through the attention mechanism, so that it can realize the target detection of cells and microorganisms. The Efficient Channel Attention (ECA) module is added to the YOLOv5 model to extract key features, and we also replace the Path Aggregation Network (PANet) of YOLOv5 with Bidirectional Feature Pyramid Network (BiFPN) for fast multi-scale feature fusion. The average precision (AP@0.5) of the improved algorithm in this paper is 81.98% under the cell and microbe microscopy datasets, which is 1.95% higher than the YOLOv5s model. The model is significantly better than the traditional deep learning algorithm, and can be effectively used for the detection of cells and microorganisms under the light microscope.
{"title":"Detection of Cells and Microbes in Microscopic Field Based on Improved YOLOv5","authors":"Xu Chu, Xiaoyang Liu","doi":"10.1109/cvidliccea56201.2022.9824054","DOIUrl":"https://doi.org/10.1109/cvidliccea56201.2022.9824054","url":null,"abstract":"The detection of cell and microbes under the microscope is of great value in both clinical experiments and experimental teaching. However, the narrow field of view of conventional light microscopes and the problem of cell or microbial stacking make target detection a challenging task. In this paper, the YOLOv5 target detection method is improved through the attention mechanism, so that it can realize the target detection of cells and microorganisms. The Efficient Channel Attention (ECA) module is added to the YOLOv5 model to extract key features, and we also replace the Path Aggregation Network (PANet) of YOLOv5 with Bidirectional Feature Pyramid Network (BiFPN) for fast multi-scale feature fusion. The average precision (AP@0.5) of the improved algorithm in this paper is 81.98% under the cell and microbe microscopy datasets, which is 1.95% higher than the YOLOv5s model. The model is significantly better than the traditional deep learning algorithm, and can be effectively used for the detection of cells and microorganisms under the light microscope.","PeriodicalId":23649,"journal":{"name":"Vision","volume":"1 1","pages":"896-899"},"PeriodicalIF":0.0,"publicationDate":"2022-05-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"90407462","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：