Capsule Defect Recognition Based On Image Processing And Classification Research

With the rapid development of modern computer technology, the developmentof image processing and pattern recognition technology has promoted quickly.Therefore, the intelligent information detection technique which based on imageprocessing and pattern recognition technology are widely used in modern industrialproduction detection.At present, the capsule detection capsule production line that mainly rely on thenaked eye detection workers, this method has great workload and high failure rate.Therefore, this paper using the modern pattern recognition technology based on theimage processing technology, a set of capsule defect developed automatic detectionsystem. This system including basic functions,such as image processing, automaticrecognition and detection function, which can detect common defects and capsuleeffectively, the difference between the qualified capsule and unqualified capsule. In theaspect of hardware, this paper uses a CMOS camera as the image acquisitionequipment,it makes use of microcomputer as the image processing equipment.This thesis studies seriously in the image processing, it analyses and compares theimage segmentation method after the image preprocessing of the capsule,thesegmentation method is proposed for an improvement watershed image. The methodlargely improved the shortage of the traditional methods of watershed segmentation，it analysis the capsule image segmentation effectively. In the detection of capsulerecognition and classification, this paper using neural network recognition technologycommon, it extracts on the capsule image feature and trains on the sample, then it usesthe improved BP neural network algorithm for the recognition and classification.This paper firstly introduced the BP neural network algorithm that is the mostcommon used pattern recognition applications. Because of the dependence on networkinitial parameters, the standard BP neural network algorithm in image classification hasslow convergence rate, which is easy to fall into local minimum problem. According tothe standard BP neural network algorithm,the paper improved three points integratedmethod improvement and excitation functions combined with the momentum factor andadaptive learning rate of BP neural network algorithm.It makes use of the improved BPneural network algorithm to detect the capsule and proves that this method is effectivefor the capsule recognition and classification through a lot of experiments.Finally,this paper implements the system in the visual programming software onVC++platform, it makes a detailed description on the overall system’s design andeach module.