Aluminum Foil Surface Defect Recognition Method Based on CNN

Abstract

In order to accurately identify the defects on the aluminum surface, including perforation, stains, yellow spots and scratches, etc., a method combining the robust principal component analysis and the convolution neural network is proposed to detect defects on the surface of aluminum foil products. Firstly, the defect area image of aluminum foil was segmented by using the method of robust principal component analysis; Then, using TensorFlow platform to build CNN network model, loading aluminum foil images for training, and save the network model parameters of training results, on the basis of this network model, loading aluminum foil images acquired in real time and classify them, complete defect detection tasks. Experimental results showed that the proposed algorithm has the following advantages such as high accuracy, favorable expansibility and so on, it can be easily applied into surface defect detection for other objects.