High-voltage transmission line fault identification is of great significance in ensuring the safe and stable operation of a power grid. This paper proposes a high-voltage transmission line fault segmentation method based on CNN-SVM. Given the complex problem of the data feature extraction process of traditional fault recognition methods, the fault features are input into convolutional and pooling layers in the form of a time series matrix through a deep learning CNN model, thereby simplifying the feature extraction and calculation process. In addition, given the problem that the fault characteristics of high-voltage transmission lines are not obvious (leading to a low recognition rate of phase-to-phase faults), it is proposed to take the current difference between the fault phases and the negative and zero sequence components of the non-fault phase as features and input them into the SVM model to determine the type of fault grounding between phases. The simulation results show that the method has a high accuracy rate. Compared with other deep learning methods, the accuracy of phase-to-phase fault recognition is improved significantly. This work is supported by the National Natural Science Foundation of China (No. 51477121).