Research On Visual Inspection Technology Of Coating Defects For Industrial Applications

Coating is a solid continuous film that acts on the surface of metal,textile,plastic and other base materials and has strengthening,protection,decoration or special functions.At present,it has been widely used in all walks of life.In the process of actual spraying and use,due to some improper operation,environmental corrosion,man-made damage and other reasons,various defects appear on the coating surface,which affects the appearance,quality and performance of the base material.In coating surface defect detection,the traditional manual detection method is too subjective,requires high experience and knowledge,and has high work intensity and low efficiency.At present,it has been gradually eliminated.The machine vision detection method based on feature description can greatly reduce the work intensity and speed up the detection speed,but its feature design is difficult,it is difficult to completely model and migrate defects,and its reusability is poor.The detection method based on deep learning has the advantages of strong learning ability,wide coverage,high application limit and good portability.It can provide an effective implementation way for the industrialized detection method of coating surface defects.This paper uses deep learning methods to conduct specific researches on the speed,accuracy,adaptability and other requirements of two typical application scenarios in the industrialized detection process of coating surface defects.The main research contents and innovations of this article are as follows:(1)This paper collects the coating surface defect data,constructs the coating surface defect data set,and designs a semi-automatic labeling tool to solve the problem of rapid acquisition of coating surface defect data in industrial application scenarios;(2)For the two typical coating surface defect detection scenarios of mobile detection and fixed detection,mobile defect detection scheme and fixed defect detection scheme are designed respectively;(3)In the mobile defect detection scheme,an improved mobilenetv2 algorithm is designed to speed up the accuracy and real-time of coating surface defect recognition;cross migration training is used to shorten the training time and improve the recognition accuracy;migration learning is used to quickly fine tune the network in the new detection environment to enhance its adaptability;(4)In the fixed defect detection scheme,the YOLOv4-tiny-SR object detection network is designed to greatly improve the defect detection speed;two optimization strategies of global geometric mean clustering and local bounding box focal loss are proposed to significantly improve the defect detection accuracy;the network parameters are quickly adjusted by using transfer learning to improve the adaptability of defect detection to the new environment.