Research On Classification,recognition And Deblurring Algorithm Of Electric Vehicle Charging Port Based On Deep Learning

With the continuous expansion of the scale of electric vehicles and the prevailing development of machine vision,the technology of using robots to automatically charge electric vehicles has developed rapidly.There were many problems in charging process,such as the uncertainty of parking location,changing weather,different light intensity,different backgrounds,changing poses of the charging port from the camera,blurry charging port images and incomplete display of the charging port in the image.Aiming at these problems,the charging port image classification algorithm,the electric vehicle charging port image recognition algorithm,the electric vehicle charging port image deblurring algorithm and the charging port coarse positioning algorithm have been studied in this article.Based on deep learning convolutional neural network,an image classification algorithm for electric vehicle charging ports was proposed.An image data set of electric vehicle charging ports under different environmental conditions was established,and training was performed on different convolutional neural network models.The accuracy of each model on the test set was compared,and the best model Inception-v3 was selected.The classification model was obtained based on Inception-v3 model training.Based on the convolutional neural network of deep learning,a target recognition algorithm for the charging port of electric vehicles was proposed.Based on the yolov5 algorithm,the complete electric vehicle charging port image data set and the incomplete electric vehicle charging port image data set under different environmental conditions were established respectively.The corresponding complete charging port recognition model and the incomplete charging port recognition model were obtained.Based on the generative confrontation network of deep learning,a deblurring algorithm for electric vehicle charging port images was proposed.The working principle and training process of the generated confrontation network were analyzed,the fuzzy image data set and the clear image data set of the electric vehicle charging port were establish,the data set was divided into six groups according to the distance between charging port area and the camera,network model training was conducted respectively and six corresponding deblurring models were obtained.Based on the PNP algorithm and the Hough line detection algorithm,a coarse positioning algorithm for the electric vehicle charging port was proposed,which can initially obtain the pose information of the electric vehicle charging port,and the robot was controlled to reach the precise positioning position combined with the control system of the robot.An experimental platform for the automatic charging and plugging system for electric vehicle charging ports was built,and classification experiments,recognition experiments,deblurring experiments,coarse positioning accuracy experiments,and plugging experiments were carried out.The classification experiments verifies that the accuracy of the classification algorithm reaches98.9%;The recognition experiment verifies that the overall recognition success rate reaches 98.7%;the deblurring experiment verifies that the deblurring algorithm can greatly improve the success rate and positioning accuracy of image coarse positioning;the coarse positioning experiment verifies that the accuracy of coarse positioning can meet the requirements of fine positioning;Finally,a plugin experiment combined with precise positioning verifies the feasibility of an electric vehicle automatic charging system.