Research On Surface Defect Detection System Of Metal Shaft Based On Improved YOLOv5

Shaft is a particularly important part of the machinery industry,almost all mechanical equipment composition can not be separated from the shaft.The existence of shaft surface defects directly affects the performance and life of mechanical equipment.Therefore,the detection of shaft surface defects is of great significance and engineering application value,however,the traditional defect detection methods all have the disadvantages of low recognition accuracy and insufficient generalization ability.The improved YOLOv5-based inspection system studied in this thesis can achieve fast and accurate detection of metal shaft surface defects.The specific research contents and results are as follows:(1)A metal shaft surface defect dataset was produced.The image acquisition platform was built,and the metal shaft surface defect images were acquired by ourselves,and the metal shaft surface defect dataset was constructed after completing the annotation task.The dataset was expanded based on common methods such as geometric transformation,optical transformation and Mosaic data enhancement.(2)A metal shaft surface defect detection model with improved YOLOv5 algorithm is constructed.Analyzing the metal shaft surface defect detection algorithm and deep learning target detection algorithm,YOLOv5 is selected as the base model and improved on it,including Anchor improvement,introduction of attention mechanism,feature pyramid improvement and other ways to improve the network model.The migration learning strategy is used in the training process of the model kind to improve the model performance.(3)Experiments of metal shaft surface defect detection based on the improved YOLOv5 model are conducted.The experimental environment of metal shaft surface defect detection with improved YOLOv5 model was built,and six improved YOLOv5 network models were obtained by combining various improvement methods,and the optimal model for metal shaft surface defect detection task was obtained by comparing ablation experiments.The experimental results show that the network models combining migration learning,K-means clustering anchor frame,CBAM attention mechanism,and BiFPN bidirectional feature pyramid improve by 3.5% compared with the pre-improvement network models,which effectively enhance the overall performance of network detection and reduce the cases of false detection and missed detection.(4)A metal shaft surface defect detection system based on PyCharm deep learning framework and PyQt5 interface design is designed for the improved YOLOv5 defect detection model and practical needs in this paper.The system visualizes the detection process of the improved algorithm proposed in this paper in detail and can switch between pictures,videos,and real-time detection,which meets the actual production needs of metal shafts.