Research Of Electronic Component Location And Identification Based On Deep Learning

Nowadays,computer vision is applied in various fields,and target recognition technology has been widely used,but the small target recognition and algorithm accuracy and speed in the application of target recognition are still to be solved.This article focuses on electronic identification problem,through the analysis of the existing shortcomings in the course of recognition algorithm in electronic components,to improve the original YOLOv3 algorithm,finally complete electronic target recognition algorithm research,electronic components and high precision real-time recognition under complex scene,laid the foundation for electronic components assembly recognition task.The main research contents of this paper are as follows:(1)Analyze the existing target recognition algorithm and the difficult problems in electronic components recognition.Firstly,the deep learning ang the detection principle of target recognition algorithm and recognition process and other basic theories are studied.Then the recognition effect of each recognition algorithm on VOC data set is analyzed,and it is found that the detection effect of existing algorithms on public data set is good,but the detection effect in complex scenes is not ideal.(2)A data set containing 9 types of electronic components was made by means of "manual acquisition+data enhancement".Then SSD algorithm and YOLOV3 algorithm with relatively good identification accuracy and speed were selected to verify the detection performance of the algorithm through this data set.For the identification of large target electronic components,the average identification accuracy of the two algorithms is 97%and 95%respectively.For the identification of small target electronic components,the average identification accuracy of YOLOV3 algorithm is 90%,while that of SSD algorithm is 88%.In addition,the recognition speed of SSD algorithm and YOLOV3 algorithm can reach 42fps and 43fps,but the accuracy of both algorithms for small targets is not very high.(3)In view of the existing in the electronic components recognition of small target recognition accuracy is not high,the YOLOv3 algorithm was improved,the first increase network scale,more characteristic of lower figure was used to extract more details information,and with characteristic figure of deeper way feature fusion based on DCA feature fusion,in order to obtain more comprehensive information;Then,the channel attention mechanism,Seblock,is added to enhance the effective feature channels in the feature layer and suppress the invalid feature channels in the feature layer,so as to improve the recognition accuracy and reduce the redundant information and improve the recognition speed.Finally,a more appropriate K value is selected based on K-means algorithm,so as to re-cluster the data set of electronic components in this paper.(4)The improved YOLOV3 algorithm was verified on the data set of electronic components.Compared with the original YOLOV3 algorithm,the MAP value of the improved algorithm increased by 0.89%and the FPS value increased by 5fps.Then,the performance of the improved algorithm was verified on the Pascal VOC dataset.Through the experimental comparison,it was found that the detection accuracy of the improved YOLOv3 algorithm was improved by 1.1%.