Features extracted by the neural network do not have scale invariance, which makes multi-scale image recognition and classification a difficult problem. Recent studies have proposed many new ways to solve this problem, such as feature fusion, sensor field transformation, etc. However, none of them essentially solve the problem that the neural network does not have scale invariance. In this paper, we propose a network generating network (NGN) architecture and design the NGNResNet network, which is an improved version of the ResNet network. The network can identify images at three scales simultaneously and has scale invariance. The experimental results show that the NGN structure helps us to improve the classification accuracy of small-scale images by about 10 percentage points, and helps to improve the performance of the network in the face of small targets.
{"title":"Network generating network for multi-scale image classification","authors":"Han Dong, Liping Xiao, Longjian Cong, Bin Zhou","doi":"10.1117/12.2671561","DOIUrl":"https://doi.org/10.1117/12.2671561","url":null,"abstract":"Features extracted by the neural network do not have scale invariance, which makes multi-scale image recognition and classification a difficult problem. Recent studies have proposed many new ways to solve this problem, such as feature fusion, sensor field transformation, etc. However, none of them essentially solve the problem that the neural network does not have scale invariance. In this paper, we propose a network generating network (NGN) architecture and design the NGNResNet network, which is an improved version of the ResNet network. The network can identify images at three scales simultaneously and has scale invariance. The experimental results show that the NGN structure helps us to improve the classification accuracy of small-scale images by about 10 percentage points, and helps to improve the performance of the network in the face of small targets.","PeriodicalId":120866,"journal":{"name":"Artificial Intelligence and Big Data Forum","volume":"32 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"123908117","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}
Stroke is a leading cause of disability in adults. Notably, about 75% of stroke survivors have upper limb damage, which greatly reduces the quality of life of the patient after recovery. The current routine rehabilitation recommendation is repetitive functional training (exercise-based training) to promote nervous system recovery, and then realize exercise rehabilitation. The cost, efficiency and success rate of traditional treatment methods are unstable due to various factors such as the professional level of therapists, the time required and the workload of therapists. In the case, rehabilitation robot-assisted therapy brings a new direction for the rehabilitation of stroke hemiplegia. In this paper, a new type of hand rehabilitation robot is designed based on the physiological structure of fingers, which is used to assist stroke patients in different stages of finger movement rehabilitation training. It can help the patient to practice grasp adduction and abduction repeatedly, reducing the burden on the patient. Secondly, in this paper, the degrees of freedom and movement of each finger joint are analyzed and calculated. Through modelling and finite element analysis based on Solid works to simulate the stress changes of exoskeleton in different rehabilitation stages, a model suitable for different stages of rehabilitation training is put forward.
{"title":"An exoskeleton rehabilitation system to train hand function after stroke","authors":"Hengyu Li","doi":"10.1117/12.2672155","DOIUrl":"https://doi.org/10.1117/12.2672155","url":null,"abstract":"Stroke is a leading cause of disability in adults. Notably, about 75% of stroke survivors have upper limb damage, which greatly reduces the quality of life of the patient after recovery. The current routine rehabilitation recommendation is repetitive functional training (exercise-based training) to promote nervous system recovery, and then realize exercise rehabilitation. The cost, efficiency and success rate of traditional treatment methods are unstable due to various factors such as the professional level of therapists, the time required and the workload of therapists. In the case, rehabilitation robot-assisted therapy brings a new direction for the rehabilitation of stroke hemiplegia. In this paper, a new type of hand rehabilitation robot is designed based on the physiological structure of fingers, which is used to assist stroke patients in different stages of finger movement rehabilitation training. It can help the patient to practice grasp adduction and abduction repeatedly, reducing the burden on the patient. Secondly, in this paper, the degrees of freedom and movement of each finger joint are analyzed and calculated. Through modelling and finite element analysis based on Solid works to simulate the stress changes of exoskeleton in different rehabilitation stages, a model suitable for different stages of rehabilitation training is put forward.","PeriodicalId":120866,"journal":{"name":"Artificial Intelligence and Big Data Forum","volume":"33 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114359209","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}
In the network geographic information system, through voice interaction, the operation can be made simple, convenient and effective. To this end, this paper studies the GIS map component technology to support voice interaction. Build the overall design of GIS map components, which includes three layers: function layer, data layer and map UI layer. The functional layer is the main layer for realizing voice interaction. After audio enters the functional layer, voice recognition must be performed first. After understanding the semantics, the mapping feedback is completed, and voice interaction is realized and supported. Experiments show that the recognition speed of the content designed in this paper is relatively fast, and the highest recognition rate is 98.5%, which provides functional component support for the information processing of geospatial information.
{"title":"Research on GIS map componentization technology supporting voice interaction","authors":"Zheng Ren, Zhen Gao, Zhengzheng Ji","doi":"10.1117/12.2672190","DOIUrl":"https://doi.org/10.1117/12.2672190","url":null,"abstract":"In the network geographic information system, through voice interaction, the operation can be made simple, convenient and effective. To this end, this paper studies the GIS map component technology to support voice interaction. Build the overall design of GIS map components, which includes three layers: function layer, data layer and map UI layer. The functional layer is the main layer for realizing voice interaction. After audio enters the functional layer, voice recognition must be performed first. After understanding the semantics, the mapping feedback is completed, and voice interaction is realized and supported. Experiments show that the recognition speed of the content designed in this paper is relatively fast, and the highest recognition rate is 98.5%, which provides functional component support for the information processing of geospatial information.","PeriodicalId":120866,"journal":{"name":"Artificial Intelligence and Big Data Forum","volume":"407 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126683533","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}
This paper focuses on the intelligent recognition of images in the Tiangong remote sensing image dataset and its interpretability analysis. In this paper, we classified the aforementioned dataset, retrained the Resnet-18 model on the training set, and then verified the results on the validation set with an accuracy of 97.9%. Furthermore, this paper presented an interpretability analysis of deep learning for intelligent recognition of the Tiangong remote sensing image dataset.
{"title":"Tiangong remote sensing natural scene intelligent recognition and interpretablity analysis","authors":"Kunnan Liu, J. Li, Guofeng Xu, Peng Wang","doi":"10.1117/12.2671376","DOIUrl":"https://doi.org/10.1117/12.2671376","url":null,"abstract":"This paper focuses on the intelligent recognition of images in the Tiangong remote sensing image dataset and its interpretability analysis. In this paper, we classified the aforementioned dataset, retrained the Resnet-18 model on the training set, and then verified the results on the validation set with an accuracy of 97.9%. Furthermore, this paper presented an interpretability analysis of deep learning for intelligent recognition of the Tiangong remote sensing image dataset.","PeriodicalId":120866,"journal":{"name":"Artificial Intelligence and Big Data Forum","volume":"37 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115621851","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}
In view of the current TCP congestion control slow-start algorithm and its waste of bandwidth due to short connections, network congestion and packet loss caused by the rapid growth of the congestion window in the later period, this paper studies the slow-start algorithm part of the TCP transmission protocol. Considering the characteristics of the current relatively high-speed network, this paper proposes an improved slow start algorithm with traffic awareness. By statistical analysis of data transmission in the network, the algorithm dynamically determines the initial congestion window size of slow start, and dynamically adjusts the congestion window by tracking the changes of real-time network traffic. In the slow start stage, the smoothness of the congestion window is further analyzed, and the smoothness of the window growth is corrected in real time, so that the congestion window does not increase exponentially, but increases by a more efficient power function. The results of this experiment show that the improved algorithm slows down the growth rate of the congestion window and improves the smoothness of the window growth. It also significantly improved the data transmission rate and throughput.
{"title":"Research on TCP congestion window smoothing control algorithm based on traffic awareness","authors":"Bing Han, Lijun Wang, Zhenliang Li","doi":"10.1117/12.2671663","DOIUrl":"https://doi.org/10.1117/12.2671663","url":null,"abstract":"In view of the current TCP congestion control slow-start algorithm and its waste of bandwidth due to short connections, network congestion and packet loss caused by the rapid growth of the congestion window in the later period, this paper studies the slow-start algorithm part of the TCP transmission protocol. Considering the characteristics of the current relatively high-speed network, this paper proposes an improved slow start algorithm with traffic awareness. By statistical analysis of data transmission in the network, the algorithm dynamically determines the initial congestion window size of slow start, and dynamically adjusts the congestion window by tracking the changes of real-time network traffic. In the slow start stage, the smoothness of the congestion window is further analyzed, and the smoothness of the window growth is corrected in real time, so that the congestion window does not increase exponentially, but increases by a more efficient power function. The results of this experiment show that the improved algorithm slows down the growth rate of the congestion window and improves the smoothness of the window growth. It also significantly improved the data transmission rate and throughput.","PeriodicalId":120866,"journal":{"name":"Artificial Intelligence and Big Data Forum","volume":"47 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115146636","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}
The rapid development of artificial intelligence has prompted the convolutional neural network (CNN) to process huge amount of data, which has caused a great burden on convolution operations. Therefore, according to the characteristics of the systolic array architecture, the acceleration structure of CNN is constructed by fusing it with CNN. Besides, it is optimized in practical application, and its effectiveness is verified. The experimental results show that in the broadcast architecture, the time required by the CNN acceleration architecture is at least 0.005, while the maximum throughput is 16.83, which is far higher than the acceleration architecture under the systolic array architecture. In the case of small change in the maximum frequency, the error rate is the same as that of the systolic array, which is about 3.62%. In the comparison of various methods proposed on the systolic array, the accuracy rate of CNN acceleration architecture is 94.7%, and the utilization rate is 81.95%. The correctness and effectiveness of the algorithm are proved. To sum up, the improved CNN acceleration structure based on pulse array optimization reduces the response time and meets the requirements of terminal calculation force, which is of high significance in practical application
{"title":"An improved CNN algorithm for accelerating structural optimization with pulsating array","authors":"Zhiliang Xiao","doi":"10.1117/12.2671338","DOIUrl":"https://doi.org/10.1117/12.2671338","url":null,"abstract":"The rapid development of artificial intelligence has prompted the convolutional neural network (CNN) to process huge amount of data, which has caused a great burden on convolution operations. Therefore, according to the characteristics of the systolic array architecture, the acceleration structure of CNN is constructed by fusing it with CNN. Besides, it is optimized in practical application, and its effectiveness is verified. The experimental results show that in the broadcast architecture, the time required by the CNN acceleration architecture is at least 0.005, while the maximum throughput is 16.83, which is far higher than the acceleration architecture under the systolic array architecture. In the case of small change in the maximum frequency, the error rate is the same as that of the systolic array, which is about 3.62%. In the comparison of various methods proposed on the systolic array, the accuracy rate of CNN acceleration architecture is 94.7%, and the utilization rate is 81.95%. The correctness and effectiveness of the algorithm are proved. To sum up, the improved CNN acceleration structure based on pulse array optimization reduces the response time and meets the requirements of terminal calculation force, which is of high significance in practical application","PeriodicalId":120866,"journal":{"name":"Artificial Intelligence and Big Data Forum","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"132711065","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}
In order to effectively supervise the wearing of safety helmets by construction personnel, the YOLOv4-tiny target detection algorithm is used to detect the wearing of safety helmets. A lightweight model with higher accuracy and less computation is designed for YOLOv4-tiny, which is more suitable for real-time helmet wearing detection. Firstly, G-Resblock is designed to replace Resblock to reduce the computational complexity of the model and occupy less computing resources. However, YOLOv4-tiny is prone to error detection or missed detection in complex work scenarios. In order to solve this problem, an attention mechanism is added to YOLOv4-tiny, the serial channel of CBAM is improved as a parallel channel, and P-CBAM is added to YOLOv4-tiny to solve the problem of poor model detection effect. The improved YOLOv4-tiny can better complete the helmet detection task.
{"title":"Research on helmet detection algorithm based on improved YOLOv4-tiny","authors":"Jianguang Zhao, Zeshan Han, Jingjing Fan, Junqiu Zhang","doi":"10.1117/12.2671490","DOIUrl":"https://doi.org/10.1117/12.2671490","url":null,"abstract":"In order to effectively supervise the wearing of safety helmets by construction personnel, the YOLOv4-tiny target detection algorithm is used to detect the wearing of safety helmets. A lightweight model with higher accuracy and less computation is designed for YOLOv4-tiny, which is more suitable for real-time helmet wearing detection. Firstly, G-Resblock is designed to replace Resblock to reduce the computational complexity of the model and occupy less computing resources. However, YOLOv4-tiny is prone to error detection or missed detection in complex work scenarios. In order to solve this problem, an attention mechanism is added to YOLOv4-tiny, the serial channel of CBAM is improved as a parallel channel, and P-CBAM is added to YOLOv4-tiny to solve the problem of poor model detection effect. The improved YOLOv4-tiny can better complete the helmet detection task.","PeriodicalId":120866,"journal":{"name":"Artificial Intelligence and Big Data Forum","volume":"68 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115735429","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}
With the development of computer network technology, the risk of network intrusion also increases greatly. But the traditional encryption and firewall technology can not meet the security needs of today. Therefore, intrusion detection technology is a new dynamic security mechanism developed rapidly in recent years. This paper studies the security mechanism used to detect and prevent system intrusion. Different from the traditional security mechanism, intrusion detection has the characteristics of intelligent monitoring, real-time detection, dynamic response and so on. In a sense, intrusion detection technology is a reasonable complement to firewall technology.
{"title":"Intrusion detection in network security","authors":"Ru-xin Wang, Yi (Estelle) Wang, Lei Dai","doi":"10.1117/12.2671429","DOIUrl":"https://doi.org/10.1117/12.2671429","url":null,"abstract":"With the development of computer network technology, the risk of network intrusion also increases greatly. But the traditional encryption and firewall technology can not meet the security needs of today. Therefore, intrusion detection technology is a new dynamic security mechanism developed rapidly in recent years. This paper studies the security mechanism used to detect and prevent system intrusion. Different from the traditional security mechanism, intrusion detection has the characteristics of intelligent monitoring, real-time detection, dynamic response and so on. In a sense, intrusion detection technology is a reasonable complement to firewall technology.","PeriodicalId":120866,"journal":{"name":"Artificial Intelligence and Big Data Forum","volume":"7 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127639095","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}
At presen, object recognition task is troubled by its huge kinds of objects. In this paper, the SIoU loss function and YOLOv5 deep learning convolutional neural network are innovatively used to improve the training efficiency and recognition accuracy. Unlike the traditional bounding box regression loss function (e.g. Giou, Diou[1] , CIoU) , which only focuses on the distance between the prediction box and the ground true box, the size of the overlap area, and one or more of the aspect ratios, and sets the impact factor on this basis, the SIoU loss function also introduces Angle cost to fit the best regression direction, which makes the direction of bounding box regression more reasonable and improves the regression efficiency[1].In this paper, the defects of traditional loss function and the calculation method of SIoU loss function are introduced, and the performance between SIoU and CIoU is compared.
{"title":"Object recognition based on improved YOLOv5","authors":"Hangong Chen, Weimin Qi","doi":"10.1117/12.2671298","DOIUrl":"https://doi.org/10.1117/12.2671298","url":null,"abstract":"At presen, object recognition task is troubled by its huge kinds of objects. In this paper, the SIoU loss function and YOLOv5 deep learning convolutional neural network are innovatively used to improve the training efficiency and recognition accuracy. Unlike the traditional bounding box regression loss function (e.g. Giou, Diou[1] , CIoU) , which only focuses on the distance between the prediction box and the ground true box, the size of the overlap area, and one or more of the aspect ratios, and sets the impact factor on this basis, the SIoU loss function also introduces Angle cost to fit the best regression direction, which makes the direction of bounding box regression more reasonable and improves the regression efficiency[1].In this paper, the defects of traditional loss function and the calculation method of SIoU loss function are introduced, and the performance between SIoU and CIoU is compared.","PeriodicalId":120866,"journal":{"name":"Artificial Intelligence and Big Data Forum","volume":"43 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116028922","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}
Based on domestic cryptographic algorithms, this research encrypts the data of multi-system collaborative IoT terminal devices, and the data is transmitted in ciphertext to realize the security protection of massive structured and unstructured data; using database encryption and decryption, file system encryption and decryption and other passwords Technology to ensure data storage and data transmission security, to achieve data confidentiality, integrity and availability.
{"title":"A data security protection mechanism for IoT terminal equipment based on multi-system collaboration based on cryptographic security module","authors":"Yali Zhang","doi":"10.1117/12.2671661","DOIUrl":"https://doi.org/10.1117/12.2671661","url":null,"abstract":"Based on domestic cryptographic algorithms, this research encrypts the data of multi-system collaborative IoT terminal devices, and the data is transmitted in ciphertext to realize the security protection of massive structured and unstructured data; using database encryption and decryption, file system encryption and decryption and other passwords Technology to ensure data storage and data transmission security, to achieve data confidentiality, integrity and availability.","PeriodicalId":120866,"journal":{"name":"Artificial Intelligence and Big Data Forum","volume":"19 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2023-03-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131269306","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：