Research On Window Button Surface Defect Detection System Based On Convolutional Neural Networ

In order to solve the problems of low efficiency,easy to miss and misdetect,this thesis proposed to use deep learning method to detect the surface defects of window buttons,and developed a window button surface defect detection system.The system is mainly composed of hardware and software.Firstly,the existing image acquisition platform with CCD camera is selected from three aspects of photosensitive element,lens and light source according to the characteristics of window button.According to the minimum defect width of the measured image,the hardware of the image acquisition platform with a CCD camera resolution of 1628*1236and a lens focal length of 35 mm can meet the image acquisition requirements of the window button under the condition of 150 mm object distance.1800 window button images were collected with this platform.Secondly,the Darknet19 network is improved by the residual block and Dropout mechanism,and the comprehensive classification accuracy of four kinds of noise for window button image is improved from 85.4% to 90.4%.Furthermore,a Do Net network with excellent generalization was proposed for denoising window button images.After the denoised images were enhanced with linear transformation,the surface defects of window button were marked with Labelme marking software,and the marked window button image data set was used to train and verify YOLOv5 network.When the crossover ratio threshold is 0.5,the trained YOLOv5 network can detect the surface defects of window buttons with72% accuracy.In addition,the hardware and software system of window button surface defect detection is built and designed.In the software system part,the Open CV driver of the image acquisition platform,the image noise classification algorithm,the image noise reduction algorithm,the linear transformation image enhancement algorithm and the YOLOv5 surface defect detection algorithm are integrated,and the window button surface defect detection operating interface is developed.Finally,the developed window button surface defect detection system is verified by image noise classification,image noise reduction,image enhancement,image detection and real-time camera detection.The actual operation results show that the software and hardware system of window button surface defect detection can realize intelligent detection of window button surface defects under experimental conditions,and the window button surface defect detection system achieves the expected goal.