Abstract: To address the challenges of drastic fluctuations in signal energy distribution and the difficulty of traditional networks in accurately capturing local energy mutation regions following short-circuit faults in ship electrical power systems, while ensuring timely response and effective handling of potential failures during maritime navigation, a short-circuit fault diagnosis method is proposed for ship electrical power systems based on empirical wavelet transform and improved EfficientNet. First, the original fault signals undergo empirical wavelet transform and fault feature visualization to generate fault feature maps. Second, the EfficientNet model is enhanced using an effective channel attention module, which strengthens its responsiveness to local high-energy regions while maintaining low parameter count and computational complexity. Finally, the fault feature maps are input into the improved EfficientNet model for training and validation to achieve accurate fault diagnosis. Experimental results show that the proposed method achieves a fault identification accuracy exceeding 99%, with significantly higher computational efficiency and resource utilization compared to other models.