A Method of Pattern Identification and Classification for Cold Rolling Steel Strip Based on Biorthogonal Wavelet Neural Network

To enhance shaping plate quality of cold rolling steel strip, a method based on biorthogonal wavelet edge extraction and neural network classification is developed. Biorthogonal wavelet filter coefficient is longer than Haar wavelet and it has more powerful anti-noise performance. Therefore, applying biorthogonal wavelet to extract target edge is first proposed to decrease data quantity used next step by neural network classification. Then the algorithm and its flowchart of neural network classification are advanced to classify the target image. At last, a field cognition system is designed to test the efficiency of this method. The results show that it can validly identify almost all defection patterns except for only two false predictions in experiment running 90 days. Compared to traditional identification system, the characteristic compression ratio and the identification precision of this method are both high. And this cognition method is also appropriate for complex steel strip pattern.