On-board cable is an important electrical equipment in electric locomotives. In order to effectively evaluate the surface hydrophobicity class (HC) of on-board cable terminals and solve the problems of low identification efficiency and strong subjectivity of manual judgment of current detection methods, an evaluation method based on multiscale retinex with color restoration (MSRCR)-spatial pyramid attention (SPA)-ResNet is proposed in this article. First, a data acquisition method of cable terminal surface hydrophobicity by spray method is proposed. Then, a data preprocessing method using MSRCR to reduce the interference of illumination angle on the feature extraction of cable terminal test image is proposed. The SPA structure is added to ResNet to form SPA-ResNet as the recognition model, which optimizes the multiscale feature extraction capability of the model and improves the recognition accuracy of similar categories. Finally, a set of evaluation methods for the surface HC of on-board cable terminals are proposed. The experimental results show that this method can effectively distinguish the HC of the on-board cable terminal surface, the recognition effect is better than the comparison scheme, and the attention characteristics of the recognition model to the test samples are basically consistent with the characteristics in the standard.