Research On Recognition And Localization Method Of Electric Vehicle Charging Interface Based On Deep Learning

As the application of electric vehicles gradually increases and their intelligence becomes more and more advanced,automatic charging technology,as the "last stop" of electric vehicle intelligence,has become a hot research topic.The technology based on the combination of machine vision and robotics is mature and widely used,which can be applied in automatic charging technology to realize intelligent unmanned and efficient automatic charging,among which the identification and positioning of EV charging interface based on machine vision plays a guiding role in this technology.At the same time,the target recognition and positioning technology based on deep learning is developing rapidly,with the characteristics of fast target detection and recognition and high positioning accuracy in general scenes,so this paper proposes a method of recognition,preliminary positioning and accurate positioning of electric vehicle charging interface based on deep learning for the current problems that charging interface recognition and positioning are easily disturbed by environment and slow recognition speed and inaccurate positioning,which realizes the The proposed method achieves the recognition and preliminary positioning of the charging interface and the accurate positioning of the charging interface contacts.The main contents of the paper are:analyzed the characteristics of electric vehicle charging interface with obvious features and easy recognition in the context of the body,proposed a feature extraction network CSP LKNet based on large convolution kernel,applied advanced deep learning tricks to improve the benchmark network Res Net,which greatly improved the speed of feature extraction without sacrificing the feature extraction capability,and provided a method for subsequent deep learning-based charging interface identification and The fast and reliable backbone feature extraction network is provided for the subsequent deep learning-based charging interface identification and precise localization methods.In the identification and preliminary localization method of charging interface,the improved identification and localization network of CSP LKNet and YOLO v5 is proposed.A charging interface dataset under actual working conditions is established,and the proposed identification and localization network is trained and performance evaluated to verify the excellent performance of the identification and localization network in terms of fast identification and accurate localization of charging interfaces.In the charging interface precision localization method,a PDNet charging interface contact precision localization network based on center point detection is proposed.The charging interface contact data set is established,and the training and performance evaluation of the precision localization network are carried out to achieve fast and accurate charging interface localization.An experimental platform was built to verify the charging interface identification and localization and precision localization methods,and experiments were conducted under different distances and light intensities.The experimental results show that the recognition success rate and preliminary localization accuracy are high in the recognition and preliminary localization experiments,and the localization accuracy reaches ±0.5mm in the precision localization experiments.the experiments confirm the reliability and feasibility of the charging interface recognition,preliminary localization and precision localization methods.