Aiming at the problem that the accuracy (ACC) value of fault line selection results is low due to the simple comparison of signal amplitude and phase in the process of fault line selection, a method of single-phase grounding fault line selection in distribution network based on signal feature extraction is proposed. S-transform strategy is developed based on the principle of continuous wavelet transform and short-time Fourier transform, and the enhancement processing of distribution network current time series signal is realized. The method of morphological transformation and morphological gradient in mathematical morphology is used to extract the mutation signal of single-phase grounding fault in distribution network, and the extracted mutation signal is used as the basis of fault line selection. The attention mechanism is introduced into the convolutional neural network, and the fault line selection model based on the improved convolutional neural network is constructed. The model parameters are adjusted through forward propagation and backward propagation training to obtain a model with optimal performance and realize accurate fault line selection. The experimental results show that under different noise interference conditions, the ACC value of the fault line selection results obtained by the proposed method is always above 0.95, which meets the requirements of single-phase grounding fault detection in distribution network.