Abstract: In the current process of identifying defects in power cables, single channel neural networks are mainly relied on to extract pulse signal features. However, the feature information captured by them is one-sided, resulting in inaccurate final defect identification results. Therefore, a multi-channel neural network-based intelligent recognition method for power cable defects is proposed. Normalize the high-frequency current signals of power cables and perform wavelet threshold denoising operations to provide high-quality data foundation for subsequent defect identification. Input the preprocessed high-frequency current signal into a multi-channel neural network, and use deep learning to extract multidimensional features such as time series, images, and statistics. Using attention mechanism to dynamically weight and fuse the features of each channel, and relying on the classification function in the fully connected layer to distinguish different types of fused features, intelligent recognition of power cable defects is achieved. The experimental results show that the F1 value of the recognition result of this method remains above 0.9, achieving accurate recognition of the defect status of power cables.