Abstract: An improved Faster R-CNN object detection model is proposed for intelligent power maintenance engineering. A dataset of power maintenance equipment was constructed, along with data cleaning and data augmentation. An attention module designed to optimize feature extraction has been incorporated into the model to enhance recognition of occluded and complex background objects. Through multi-scale processing and cross-space learning methods, the model is able to capture and utilize critical information in the image features while maintaining computational efficiency, thereby improving overall performance. Several experiments were designed and the model effects visualized. The experimental results further validate the effectiveness of the model. This research holds theoretical significance and practical value for enhancing the efficiency and quality of power maintenance work, ensuring the safe operation of power systems, and supporting the realization of the "Dual Carbon" goals.