Research On Wheel Surface Defect Detection Based On Improved Yolov4 Algorithm

With the strategy of "Made in China 2025",China’s manufacturing industry is in the transition stage of intelligent development,and the automotive industry,as an important part of the national economy,is a crucial part of the attack on intelligence.As a key component of the automobile,the wheel hub may produce many defects such as burr,porosity,sticky aluminum and so on in the actual production and processing,which has a great impact on the appearance of the automobile and the performance of the whole vehicle.The traditional wheel defect detection mostly adopts manual visual inspection,which is time-consuming and labor-intensive,and it is very easy to cause mis-inspection and leakage due to worker fatigue,and the accuracy is difficult to guarantee.With the continuous development of computer vision technology,vision-based defect detection methods are more and more widely used.This paper is based on computer vision technology to detect defects on the wheel surface in the wheel production line in order to improve the accuracy and efficiency of wheel defect detection and further promote the intelligence of the wheel production line.The main research contents of the paper are as follows.(1)The data set,as a prerequisite for intelligent detection,is the primary problem studied in this paper.Aiming at the wheel defect images collected from industrial sites,we determine the types of wheel defects,analyze the quality and quantity problems of wheel defect images,solve them with reasonable image processing means,complete the annotation of wheel defect data with the help of annotation tools,and realize the construction of wheel defect datasets.(2)The structure of the Yolov4 algorithm and its innovative ideas are analyzed,the algorithm is pre-trained using the wheel defect dataset,the detection results are initially obtained and analyzed,and the optimization direction of the algorithm is clarified based on the characteristics of the detection result data and the defect dataset itself.(3)Research the improvement ideas based on the Yolov4 algorithm,construct the network structure of the improved Yolov4 algorithm;design ablation experiments and comparative experiments,and verify the innovation and superiority of the improved Yolov4 algorithm proposed in this paper according to the detection accuracy and detection efficiency.(4)Based on the above algorithm,an on-line detection system for wheel hub defects is designed and developed that can realize real-time detection and complete a series of operations such as display,marking,saving,and statistics.