Defect Detection System Of Cloth Based On Convolutional Neural Network

A defect detection algorithm of cloth based on Neural Network by involving effective use of image processing and neural network is presented in this paper. The samples collected on the surface of the cloth are preprocessed by wavelet transforming and Otsu method, then they would be identified and classified through AlexNet. The defect information on the surface of samples is removed by filtering, and the feature is strengthened by threshold method. The image is adjusted to meet the requirement of neural network. The training data is learned by the feature detection layer, so as to achieve the detection of test data. It can detect the flaws on the cloth fast and correctly, and raise the product quality and improve production efficiency. Through the study of 400 collected samples, this method is applied to the 40 samples for testing. The success rate of the trained neural network is 99.2%, and the actual test accuracy was 93.33%, which is higher than 81.8% of Gabor method, 87.2% of MRF method and 90.4% of SE algorithm. It is considered as a suitable way for flaw detection and has a good application prospect.