In the realm of conventional substation inspections, delays in problem detection or inaccuracies in data recording by inspection personnel may arise due to various factors. The application of modern technologies such as three-dimensional reconstruction, augmented reality, and deep learning proves instrumental in mitigating these challenges. Initially, a comprehensive and detailed modeling of the substation is achieved through the use of a three-dimensional laser scanner. This model, augmented with colored textures derived from aerial images captured by drones, forms a three-dimensional representation of the substation, which is then uploaded to the cloud. Addressing issues such as the assessment of the main transformer's breathing apparatus status and identification of foreign objects within the equipment, deep learning methods including HSV (Hue, Saturation, and Value) color channel segmentation and YOLOv5 are employed. This leads to the development of algorithms for evaluating the main transformer's breathing apparatus status and recognizing equipment anomalies, deployed on the cloud to establish a cloud-based digital twin system. Operational personnel, equipped with augmented reality glasses during inspections, engage in real-time interaction with the cloud. This interaction yields status information and historical data for the inspected equipment, with the option of remote expert support when necessary. Furthermore, remote monitoring of inspection personnel status in real-time, facilitated through the augmented reality glasses, enables timely warnings of high-risk operations. This integration significantly enhances the efficiency of substation inspections, reduces error rates, minimizes the likelihood of accidents, and ultimately reinforces the reliability of power supply.