The Algorithm Research Based On LBP For Photoelectric Detection System Of Bullet Surface-defect

The bullet surface-defects automatic photoelectric detection equipment has a great application prospect of the military industrial enterprises and defense agencies in our country, but the automation of bullet defect detection is not well applied, one of the most important reasons is that the defect detection algorithm is not up to par. Because of the curving surface and the particularity of metal material, the low contrast between the background and the defects is caused in the bullet images, and usually having uneven illumination at the same time. In this situation it greatly increased the difficulty of detection algorithm.Texture analysis method is one of the most effective image processing methods springing up in recent years. In the texture analysis family, local binary pattern being proposed so far has been widely studied and applied in many fields, such as face recognition, visual inspection, etc., and shows its extraordinary advantages. Considering the excellent characteristics LBP shows and the processing difficulty which bullet images have, applying LBP into bullet detect defection is an innovative attempt, and it has been deeply discussed in this dissertation.At first, the definition and classification of defects is the basis for defect detection.Relevant knowledges about bullet is studied, defect category is divided, and bullet images(including the defect ones and the qualified ones) are obtained through the bullet photoelectric image acquiring system, then, the characteristics of the images are analyzed. Defect detection process is developed after carefully analysis of the bullet images’ characteristics, then, preprocessing program of de-noising and extraction of the region of interest is proposed.In the stage of feature extraction, various features of LBP operator are investigated through learning its definition and calculation method. Three kinds of LBP derivative operators which can be used in bullet surface-defect detection are chose considering the features of the actual bullet images. These three methods are deeply discussed in this thesis, and according to the experiment results of bullet surface-defect extraction, finally the most suitable operator LBPV is chose out for the research of bullet surface-detect detection.Based on LBPV operator extracting texture features, method and algorithm whichcan be adapted to bullet image segmentation and defect recognition are studied, in the result, bullet defect recognition algorithm which combined with FCM is proposed.Further classification of defects is studied on the basis of the recognition of defects, and in result bullet defect classification algorithm which combined with K-nearest neighbor classifier is proposed.The experiment results show that the bullet defect texture recognition algorithm based on LBP can effectively eliminate the serious effect of uneven illumination for defect detection, and small defects can be successfully detected, miss rate is reduced to1.9% and accuracy is above 97%. As far as we know, it is the first report in China.Experiments prove that this algorithm can be successfully used for surface-defect detection of bullets, at the same time it can also be used to surface-defect detection of other metal objects.