Research On Some Problems About Recognition Of Surface Defects In Steel Plate Based On Neural Network

Nowadays, steel companies generally adopt the behindhand manual visual method and strobe light detection for the quality of steel plate. However, these methods have many shortcomings, such as low inspection percentage, bad real-time performance, low detected degree of confidence, formidable circumstance of detection, etc. Therefore, the research about on-line nondestructive detection of defects in steel plate has become a topic concerned by scholars and suppliers of automatic detection equipment around the world. At present, artificial intelligence diagnostic system based on neural network is one of the hot items. It indicates its unique superiority in many aspects of pattern recognition, procedure mode building, control, fault diagnosis, etc, because it has many characteristics such as massive parallel process, distributed storage, learning capability and so on. Consequently, neural network has widely application foreground in recognition of defects in steel plate. This thesis analyses its development state at home and abroad, attempts various methods of image processing, explores related theories and key components, and successfully applies neural network method into defect recognition. The major tasks and achievements are as follows:The first is design of system. The paper analyses quality detection system of steel plate, and sets up complete hardware system and software process. Against the productive circumstance of steel plate, the author proposes relevant opinions on detection light source and CCD camera. The paper presents processing and recognition course according to the characteristic of defects in the steel plate.The second is image processing. The paper adopts a self-adaptive weight averaging method against the shortcomings of mid-value filtering, which can largely improve effectivity of filtering and assure the speed simultaneously by introducing noise detecting and reserving most of details. Small wave filtering is put forward by introducing Laplace operator to image processing, which enhances adjustability of image filtering by combining sharpening with smoothing. In order to improve limitation of classical Canny operator, the author considers diagonal direction of pixels and leads it into defference averaging calculation, which improves accuracy of edge locating, rejects noise, and increases its effect.The third is extraction of feature. The paper brings forward an improved method which extracts projection feature parameters by proving its adaptability in defferent axes of rotating angles. Furthermore, the author introduces Hu invariant moment and Zernike moment, which...