Infrared detection of substation equipment can detect overheating defects effectively in time, which can prevent the occurrence of equipment faults. It is difficult to analyze and process massive infrared images from autonomous inspection of substation robots and unmanned aerial vehicles(UAVs) with traditional manual diagnosis method. However, at present the intelligent diagnosis based on infrared image is mostly based on traditional machine learning algorithm, which has low recognition accuracy and poor generalization ability. Firstly, object detection of transformer, bushing, circuit breaker and other seven substation equipments is carried out based on improved Faster RCNN method, which realizes accurate positioning and identification of equipment. Then, defect recognition is carried out based on temperature discrimination method. The infrared image collected from the scene is used for testing, and the recognition accuracy of the seven types of equipments reaches more than 90.61%, and the accuracy rate of defect recognition accuracy is 81.33%. The experimental results show the effectiveness and accuracy of the proposed method.