Research On Defect Detection Algorithms For Steel Plate Surface Based On Deep Learning

Steel is the backbone of a strong nation,and the surface quality of steel plates,an essential raw material for industrial production,directly affects the quality of the end product.The standardisation of steel plate quality is a must for the Chinese manufacturing industry to move towards the middle and high end,and the existing methods of steel plate surface defect detection are becoming increasingly difficult to meet the needs of enterprises,so it is of great relevance to carry out research work in the field of steel plate surface defect detection.To address the contradictory phenomenon that the current common steel plate surface defect detection algorithms are difficult to meet the actual production needs in terms of practicality and real-time indexes,this paper integrates the knowledge of image processing and deep learning to improve and build a lightweight network classification model represented by Mobile Net V3 and a target detection model represented by PP-YOLO with the task of classifying and detecting steel plate surface defects as the target.detection model represented by Mobile Net V3 and the target detection model represented by PP-YOLO,and scientifically introduce relevant evaluation metrics for analysis.The paper firstly compares the architecture and feature knowledge of the convolutional neural network involved in this task,and summarises the advantages and disadvantages;then,to address the phenomenon of overfitting arising from the model learning limited by small samples,the defective samples are expanded by using data enhancement methods such as translation,rotation and noise addition,while three filtering algorithms,namely mean,bilateral and median,are designed to obtain clearer images of the defective samples,and For the classification detection task,the original Mobile Net V3 model is improved from the perspective of structure and parameter quantization,and the model converges faster and is more robust compared to the lightweight models represented by Mobile Net V2,Shuffle Net V2,and the knowledge distilled Res Net50;finally,for the target detection task,the The four groups of models,YOLOV3,YOLOV3-Mobile Net V1,PP-YOLO and YOLOV4-Tiny,were tested for comparison.The evaluation parameters show that the PPYOLO model is the best in terms of inference speed,recognition accuracy and other indicators,and has good generalization ability in the task of steel plate surface defect detection.The paper finalises the research and validation task of the deep learning-based surface defect detection algorithm for steel plates,and performs tests of the expected functionality and results of each algorithm.The experimental results show that the image processing algorithms,image enhancement schemes,classification and detection models selected and designed in this paper are able to perform the task of detecting surface defects on steel plates very well.