This study presents the design and implementation of an intelligent diagnostic system for power equipment status based on infrared imagery, leveraging deep learning techniques to enable automatic fault detection and localization. The approach utilizes the Mask R-CNN model to process infrared images, integrating image pre-normalization, feature extraction, and data augmentation strategies to enhance model reliability and robustness. Experimental evaluation demonstrates that the Mask R-CNN model achieves an average precision of 92.3%, an identification accuracy of 94.5%, a precision score of 93.8%, and a recall rate of 90.6% in detecting power equipment faults, outperforming baseline models such as Faster R-CNN and YOLOv4. The results verify that the deep learning-based diagnostic system exhibits high efficiency and accuracy in fault identification, significantly contributing to the safety and operational stability of power systems.