|
This paper uses factory-made glass sheets to simulate the final appearance inspection of the glass production process. The surface of the glass sheets may have different defects such as bumps, bumps, cracks, etc. The glass has light reflection characteristics. Traditionally, due to the artificial The comparison detection method is not only time-consuming and labor-intensive, but also prone to different detection results due to the interference of light reflection, resulting in the outflow of defective products. Recently, the topic of deep learning has become more and more discussed, and it is also widely used in image recognition and image recognition. In this study, deep learning is used to mark and train the defects of glass sheets in the image, and the detection frame of defects on the glass is selected by the method of object detection, which can reduce the misjudgment rate and the occurrence of defective product outflow. Can improve detection efficiency. The detection equipment in this research uses Raspberry pi 4, and uses the popular object detection algorithm YOLOv5 for glass defect detection. First, hundreds of glass defect images are collected, and the image defects are marked by the frame selection tool labelimg Relative position, save the marked data and images to the YOLOv5 data set, and then adjust the program inside the data set. After adjusting the program, execute the command on the terminal to test the glass defect detection effect. In order to effectively improve the detection of glass defects For the efficiency and accuracy, it is necessary to repeatedly execute the observation and detection results to adjust the command to find the best object detection training set. Through countless training and parameter adjustments, although the results of glass flaw detection are not ideal, there are also detections of glass scratch with a 98% recognition success rate. Through this research, deep learning can indeed Items are inspected for defects, which also proves that this application is feasible in industrial appearance inspection.
|