Steel Strip Surface Defects Recognition Based On Image Analysis

Based on the analysis of demand on video monitoring of strip rolling line of steel company,this thesis applies the theories to the algorithms to get better results.This scheme includes three process of disposal such as image pretreatment,feature extracting,classifying.Aiming at the fault diagnosis for steel strip,an on-line real time fault diagnosis system is developed.As a result,the percentage of up to standard products was greatly enhanced.At first this thesis expounds the digital ima ge preprocessing and develops an image enhancement algorithm of histogram equalization.The flourier transform cannot perform a good effect to decrease the interfering noise.In order to filter the noise in image,it presents an active filter b ased on adaptive filter theory.Segmentation of image is an important aspect in analysis and recognition.In order to remove the speckle noise,the thesis processes the image with a morphology filter,and then segments it with Markov random fields,Gibbs random fields and maximum a posteriori probability,finally extracts character with though transform.This method can give a best threshold for each image to obtain a good impact of image segmentation.It describes an extraction method of texture feature,shape feature and spatial feature.It presented a method of image classification using texture feature,Hu invariant moments and dispersion degree.Two new recognition approaches based on genetic algorithm and particle swarm are proposed.Experimental results show that the pso is an effective method in solving the problem compared to the same subjects.The Two kinds of algorithms have their advantages and disadvantages.