| Defects will inevitably occur in the industrial production process.In order to avoid the low efficiency of traditional manual inspection,automatic defect detection based on machine vision is widely used in industrial fields,and the detection tasks of machine vision require high resolution and real-time processing.At the same time,the rapid development of deep learning technology has also gradually been applied to object detection and achieved good results,the application of deep learning technology for defect detection in the industrial environment has developed slowly.For this reason,this thesis deeply researches defect detection algorithms in the industrial environment,and analyzes the reasons for the slow development of deep learning technology in this field.Then,in order to ensure the quality of the product and realize rapid defect detection of the workpiece,guided by practical application projects,this thesis conducts research on workpiece defect detection algorithms in an industrial environment for the characteristics of different workpieces,the main work is as follows:Based on the analysis of the existing machine vision based defect detection algorithm implementation process,for the complex flat-ear ear defect detection task,this thesis proposes a flat-ear ear defect detection algorithm combining the template matching algorithm and the corresponding relationship of the defect position.On the industrial dataset,the detection rate of the algorithm is 96.8%,and false alarm rate is 4.2%.For a challenging ultra-high resolution mobile phone screen defect detection task,this thesis proposes a defect region selection algorithm based on machine vision.When the algorithm processes image data up to 8192×11000 pixels,the average time to complete the production of candidate regions is only 382.27 ms.Finally,this thesis proposes an end-to-end mobile phone screen defect detection algorithm based on deep learning which combining the defect region selection algorithm and YOLOv3 algorithm.For the ultra-high resolution mobile phone screen defect detection task,the detection rate of the algorithm is 96.8%,and false alarm rate is 4.2%. |
