Abstract: In this paper, the YOLO V5 target recognition algorithm of the depth learning technology was used to identify the leakage region of the infrared imager from collected infrared images. For infrared leaky targets, the sample size of infrared leaky targets with different background conditions and different time was small and difficult to collect, which made the training of deep learning model very difficult. In order to reduce the need of real infrared leakage image, this paper combined simulated infrared leakage image with real infrared leakage image to make data set, train these data set as a sample for deep learning. The experimental results showed that the accuracy of the proposed method was 87. 6% for identifying the leakage area in the infrared image of the building facade.