The invention discloses a transmission line fault current traveling wave feature extraction method, which relates to the field of transmission network fault detection. The traveling wave signals at the transmission line fault moment are acquired and integrated, normalization processing is carried out, multiple convolutions and pooling calculations are carried out, the original waveform features are enhanced, the noise and data dimension is reduced, and the visual consistency of the traveling waves and the features is kept. Besides, the value and the length of the convolution kernel in each layer are different from those in the other layers, each convolution kernel slides in the entire signal, the same feature of the image at different positions is detected, the final convolution results constitute a feature map of input signals, and the local features of the signals are extracted; the features are then compressed by pooled downsampling, the main features are retained, the data dimension is reduced, the calculation processing amount is reduced, the calculation speed is improved, the consistency between the original traveling waves and the visual features is better kept, and the recognition accuracy is higher.