| The surface defects of chip packages would worsen the performance and reliability of chips.In order to achieve the detection of various types of ones and improve the quality of packages for chips,it is necessary to focus on the detection techniques of surface defects.Based on the photometric stereo vision technology,the detection algorithm for micro-defects on chips surface is designed as well as a detection system for those is implemented with accuracy of 0.04 mm.The main research content is as follows:(1)A hardware prototype system was designed and built to collect chip images.In order to build a better hardware prototype system,comparison experiments with different numbers of light source and corresponding angles were carried out.And a 4-light source photometric three-dimensional scheme suitable for chip appearance image detection was designed.It can help to get high quality photometric stereo images,and not worsen the real-time by the scheme.(2)An image enhancement algorithm was proposed.The image quality obtained by the photometric stereo algorithm is much higher than that of the image collected under a single light source,but the features of small defects cannot be accurately captured,resulting in low accuracy of defect detection.The Gabor transform can simultaneously obtain the characteristics of the image signal in the time domain and the space domain for image enhancement,and can effectively obtain the local microdefect characteristics of the chip image.An image enhancement algorithm combining photometric stereo technique and Gabor transform was designed,which improved the detection accuracy greatly to0.04 mm for small defects.(3)A defect detection algorithm based on YOLOv4 was put.Firstly,the high-dimensional reconstruction of the image was carried out according to the gray-scale characteristics of the chip image.The purpose was to highlight the information characteristics of the entire image,while highlighting the subtle defect features,which was conducive to subsequent defect detection and recognition.Then,the YOLOv4 was trained by using the grayscale 3D reconstruction image as the model training sample,and expected to realize the classification and location of defects suitable for a variety of chip types.The experimental results showed that the small defect features on the grayscale 3D reconstruction image can be accurately extracted under the YOLOv4 target detection model.(4)A chip surface defects detection system was designed and implemented basing on the key algorithm mentioned above.The overall software architecture was constructed according to actual industrial application requirements by field research.Finally,a set of chip defect detection system,which was with smooth human-computer interaction and high system accuracy,was developed.This thesis has made a certain technical exploration on the detection technique for surface defects on chip packages.And it has made some technical accumulation for the subsequent developing of a more optimized high-end chip appearance defect detection system for enterprises.The work of the thesis can give some reference to someone who would like to explore stereo vision technology and deep learning technology in the field of chip appearance defect detection!... |
PDF Full Text Request