The Research On Strip Surface Defect Images Detection Based On Visual Attention Mechanism And D-FNN Recognition

The strip is the indispensable raw materials in aviation,automobile and chemical industry.In recent years,with the development of technology and the increase of the quantity demanded for strip steel,the requirement of strip surface quality is increased,However,but due to the level of technology and production equipment,its surface will appear such as black spots,scratches,warping,phosphating spots,wrinkles and other defects in the process of rolling strip,these defects can not only affect the appearance of the product,but also can affect the performance of the product and it will bring unexpected security risks for consumers and producers,so the research of application of machine vision detection system on the strip surface image has important significance.In this paper,the key points of strip surface defect image detection and recognition are analyzed.The effective algorithm is obtained by comparing the results of experiments.The research content and achievements in this paper mainly includes the following aspects:1.A method of strip surface image denoising based on BM3 D is studied and it describes the working principle of the algorithm.After several contract experiments,the optimum parameters for defect image denoising is obtained by BM3 D algorithm.The comparison of the BM3 D algorithm with setting parameter values and several classical denoising algorithms show that whether from the subjective judgment standard or the objective judgment standard,the BM3 D algorithm has obtained the better effect in the image denoising aspect.2.According to the variety,complex shape and random distribution of strip surface defects,the algorithm based on the Itti model of the visual attention mechanism is studied and the values of feature weights of the algorithm are defined by the saliency entropy of each feature map and the Itti algorithm is used in the image area of the strip surface defects segmentation,what's more,the WTA(winner-take-all)network mechanism is used to locate the position of the defect region in the image.Finally,compared with several classical segmentation algorithms,the experimental results show that the Itti algorithm has achieved better results.3.By extracting the values of characteristics of the surface defect image of geometric features,gray feature,invariant features and topological features,texture features,used the D-FNN(Dynamic Fuzzy Neural Network)algorithm to identify the surface defects of steel strip.compared with several classical neural network algorithm(BP neural networkalgorithm,RBF neural network algorithm),the experiments show that the D-FNN algorithm has better effect in the defect image recognition.