Research On Neural-Network-Based Flaws Classifier Algorithm For PS Plate

With the development of NN and related technology, the classification and measure method has been more and more used in the detection of complex material. In the presensitized plate surface defect detection, in order to increase production line speed and accuracy of defect detection and defect production lines to achieve real-time control, machine vision is better than the human in terms of cost and speed. Using machine vision technology for surface defect detection and classification, the features are a detection accuracy, a wide range of defect recognition, high reliability and efficient, It is a hot point in the surface defect nondestructive defection of research field. According to the characteristics of presensitized plate for its surface defect detection, this paper presents the improved defect identification method based on a BP neural network, at first do the unitary dispose of the image after match, avoid the external environment impact to the image, then the image is divided into sub-blocks, sub-block extracting the gradient histogram as the feature vectors, these extracted feature vectors input of neural network. The BP neural network is one of the latest developed and the most widely used classification methods. It is easy to avoid the other neural network more learning problems and effectively solve the classification problem.Aiming at the shortcomings of the BP algorithm, we improve it by dynamically changing the learning speed and adding momentum factors. This paper describes design and implementation of the fabric flaws classifier in detail. The results shows that the current fabric flaws classifier has got a completeness rate and an accuracy rate.