Semantic segmentation model training method, device and equipment based on context awareness

The invention provides a semantic segmentation model training method, device and equipment based on context awareness. On one hand, a full-size remote sensing image is cut to obtain a high-resolution local image, and then the high-resolution image is subjected to down-sampling to obtain a context cut image, so that the capability of capturing context information of the remote sensing image by a model can be trained based on the part of the image; on the other hand, the high-resolution local image is further cut, so that the detailed part of the remote sensing image can be accurately segmented based on the image encouraging model; and finally, the prediction results with different resolutions are aligned and fused to obtain a final prediction result, so that the capability of the model for identifying context information and detail information of the remote sensing image can be effectively trained, and the final segmentation precision of the model is improved. Besides, the random mask technology can be utilize