Remote sensing image building detection method and system based on ResUNet

The invention discloses a remote sensing image building detection method and system based on ResUNet, and the method comprises the steps: constructing a U-Net building extraction network of double attention residual errors, segmenting an input remote sensing image into binary building maps, binding any polygon in each building map with a bounding box, so as to filter non-building frames with imperfect building extraction; comparing the corresponding construction frames in the biphase building map to obtain a building change map, evaluating the training performance of the comparison model to obtain a building extraction result, detecting the building extraction result to complete building change detection, increasing the attention of the model to construction of high-level semantic information, and improving the construction efficiency. A channel space attention residual error convolution block replaces a U-Net simple convolution block, the attention degree of a model on valuable building bottom layer space fe