Application of a convolutional neural network in wire rope magnetic memory testing

Abstract

In this paper, a magnetic memory detection device under weak magnetic field excitation is designed to better solve the problem of weak magnetic memory detection signals and susceptibility to other factors. In order to reduce the noise in the original signal, a noise reduction method combining local mean decomposition and wavelet transform (LMDW) is proposed. Pseudo-colour transformation is used to enhance the greyscale image after cubic spline interpolation. Finally, a convolutional neural network (CNN) is designed to identify broken wire. Moreover, compared with the support vector machine (SVM) algorithm, the recognition rate of the CNN is 35.8% higher than that of the SVM under the condition that the allowable error is 0. The experimental results show that the system has high detection sensitivity and remains effective for small defects. The filtering algorithm has a better effect on noise removal and improves the signal-to-noise ratio (SNR). The CNN has good recognition ability to identify defects.