In order to solve the problem of dam leakage detection , this paper presents a dam leakage detection method which combines active excitation infrared imaging with depth learning. The infrared image of leakage is pro- duced by computer simulation , and then combined with the infrared image acquired by simulating dam leakage test , the infrared image data set is generated for the training of depth learning. On the basis of Yolov5 original model , using AF-FPN to replace FPN can improve the ability of identifying the leakage area of dam infrared image , and make an ef- fective trade - off between identifying speed and accuracy. The test results show that the accuracy of the model is 87. 6% , the recall rate is 96. 5% , and the average accuracy ( IoU = 0. 5) is 88. 3% , which indicates that the method proposed in this paper can identify the leakage area of dam infrared image well.