SAH-Unet-based high-resolution remote sensing image impervious surface extraction method

The invention discloses a high-resolution remote sensing image impervious surface extraction method based on SAH-Unet, and the method comprises the steps: S1, obtaining a target region high-resolution remote sensing image and corresponding OSM data, and carrying out the labeling of the high-resolution remote sensing image, and obtaining an impervious surface label image; s2, preprocessing data to obtain an impervious surface sample data set; s3, constructing an SAH-Unet model for feature extraction of the high-resolution remote sensing image; s4, taking the impervious surface sample data set of the target area obtained in the step S2 as network input, and carrying out iterative optimization on SAH-Unet model parameters by using a neural network optimizer through minimizing a loss function; and S5, inputting a target high-resolution remote sensing image of a to-be-identified impervious surface into the SAH-Unet model, extracting image fusion features and performing pixel-by-pixel surface feature category predi