Surface Defect Detection Based On Deep Learning Methods

With the development of artificial intelligence technology and the popularity of intelligent production projects,automated artificial intelligence systems have gradually become a hot research topic in the industrial field.Object detection,as a fundamental issue in the field of computer vision,is still an important challenge in the development of automated object detection systems,especially when the accuracy and realtime performance of the detector need to be balanced.To this end,this paper introduces the MSFT-YOLO model,which is improved based on the one-stage detector and designs a material surface defect detection system based on this model,with the following details and results:A systematic introduction to current object detection and material surface defect detection methods,detailing their basic principles,model structures,and applicable scenarios.Experimental studies and comparative analyses of commonly used one-stage detectors and two-stage detectors were conducted,by weighing the usage conditions and performance of the model,the direction of the algorithm framework selection and the main requirements of the system are clarified.Developed an improved algorithm scheme based on the one-stage detector,which is suitable for material surface defect detection tasks.The MSFT-YOLO model is proposed for the industrial scenario,in which the image background interference is great,the defect category is easily confused,the defect scale changes a lot,and the detection results of small defects are poor.By adding TRANS module based on Transformer design to the backbone and detection headers,the features can be combined with global information;The fusion of features at different scales by combining BiFPN structures enhances the dynamic adjustment of the detector to targets at different scales.With the MSFTYOLO inspection model as the core,a B/S architecture-based online inspection system for material surface defects is built.With the aim of enhancing user experience and improving platform detection efficiency,using a variety of development tools such as Element-UI,Axios and MongoDB,an interactive and stable online inspection system for material surface defects is designed.Also,a simple-looking interactive UI interface is realized through modular design and card layout,which greatly improves the convenience of using the system.Integrated login verification,system settings and other basic functions and uploading pictures for detection,detection effect feedback and other system functions to achieve intelligent online detection.Comparison experiments were designed to verify the effectiveness of the algorithm.Experimental results show that the model can achieve high detection accuracy while also having the ability to detect in real-time.The test results on the NEU-DET dataset show that the mAP of MSFT-YOLO is 75.2,improving about 7%compared to the baseline model(YOLOv5),which can solve the problem of poor detection of material surface defects in industrial scenes for images with strong background interference,large changes in defect scale,a large number of small defects and easily confused defect targets.Also combined with the browser,Postman and other third-party software on the material surface defects online inspection system for functional testing,the platform test results show that the core modules of the system function correctly and operate well,which have high engineering application value.