Abstract: To solve the problem of surface defect detection in inaccessible parts of large-scale special equipment, a method using unmanned aerial vehicle (UAV) to detect and identify surface cracks was proposed. Firstly, a UAV detection device equipped with a dual pan-tilt-zoom (PTZ) platform was used to comprehensively collect surface images of the tank farm cofferdam walls and high-altitude building walls; Then, the Faster R-CNN deep learning neural network algorithm was used to classify the collected images and detect whether there were cracks or defects in the images; Finally, morphological processing on the detected crack target box area was performed. The detection results showed that the Faster R-CNN algorithm had a crack detection accuracy of 95.74%, with a crack width recognition error of about 3.9% and a length error of about 5.3%. It had achieved remote automated detection of the tank farm cofferdam wall and high-altitude building wall.