With the expansion and complexity of the power system, the safety monitoring of the substation operation environment has become a key link to ensure the life safety of power operators and the stable operation of equipment. This study proposes a safety monitoring method for substation operation that integrates high resolution remote sensing technology and improved YOLOv5 deep learning model, and constructs a complete monitoring system through three core modules: remote sensing image preprocessing, target detection algorithm optimization about safety equipment and dynamic detection of dangerous areas. In remote sensing image segmentation, a multithreshold segmentation method is used to eliminate geometric distortion and radiation distortion and extract key feature information. For the problem of small target detection in complex scenes, the YOLOv5 model is improved, the coordinate attention mechanism CA is embedded to enhance the feature extraction capability, and the SPPF module is reconstructed by using the large kernel separated attention convolution, which is combined with the GD aggregation-distribution mechanism to optimize the necking network and to improve the multi-scale target detection accuracy. The experimental results show that the improved model has a denoising performance PSNR of 28.98dB, an SSIM of 0.874, and a safety equipment detection F1 value of 95.10% for insulated suits and 95.55% for safety helmets. The average accuracy of hazardous area misentry detection is 95.85%, etc. are significantly better than YOLOv3, Faster RCNN and other comparative models, and the computational efficiency is high, and the detection speed reaches 7.26ms/mg.
