Abstract: In this thesis, adopting the character of grey histogram, grey level co-occurrence matrix and wavelet transform, three kinds of characteristic methods combineing can the purpose of the good realization classification.On the foundation of the feature extraction, this study accord to environment of MATLAB6.5 under of the nerve network tool box, RBF nerve network was put forward, which consider both the speed of recognition and the classification accuracy to recognize and classify the strip steel surface, This arithmetic can be the optimization project tothe real-time examining of the high-speed production line of strip steel surface defect image.

Key words: Matalab, RBF neural network, BP neural network, strip steel defect image, discern and classification, feature extraction

摘要： 本论文中，采用灰度直方图特征、灰度共生矩阵特征和小波变换特征的提取方法，三种特征方法的结合能够很好的实现分类的目的。在提取特征向量的基础上，本研究基于MATLAB6.5环境下的神经网络工具箱，采用了兼顾识别速度与分类准确性的RBF神经网络分类器对带钢表面缺陷进行识别与分类，此算法可以作为高速生产线的带钢表面缺陷的实时检测优选方案。

关键词: Matalab, RBF神经网络, BP神经网络, 带钢表面缺陷, 识别与分类, 特征提取