The image background of autoclaved aerated concrete block prefabricated multi story buildings includes different building materials， colors， textures， etc. These background information results in insignificant crack features of different shapes， leading to poor crack recognition performance. Based on this， a crack identification method of autoclaved aerated concrete block fabricated multi-storey building based on SR-UNet is proposed. Lucas Kanade method is introduced to track the feature points of crack images and determine the specific positions of crack feature points in the current frame image. SR-UNet network segmentation model is built to segment images containing crack feature points， and process the input images through dense residual connection modules and attention modules to obtain segmented feature images. By using boundary loss and region generalized Dice loss， an overall loss function is formed， the difference between the segmented image and the real image is calculated， and the model parameters are optimized using backpropagation algorithm to achieve the recognition of cracks in prefabricated multi story buildings. The experimental results show that the proposed method has the highest recognition accuracy of 98% for cracks in prefabricated multi story buildings， which is superior to the three compared methods. Moreover， it can fully recognize crack images in low light environments and small crack images.