Training method, device and equipment of semantic segmentation model based on boundary enhancement

The invention provides a training method, device and equipment of a semantic segmentation model based on boundary enhancement. According to the method, boundary detection can be realized by extracting the boundary features of the remote sensing image, and the boundary features can be utilized to enhance the multi-level features containing the context information and guide the semantic context, so that the model can have a more detailed context view angle, and the semantic segmentation performance of the model can be improved. Besides, when multi-scale semantic information is extracted and boundary information is used for guiding multi-level semantic feature fusion, only simple parameter-free mathematical operation is used for fusing two complementary features, the fusion performance is guaranteed, meanwhile, the complexity is reduced, and the parameter quantity is reduced. 本申请提供一种基于边界增强的语义分割模型的训练方法、装置和设备。其中，通过提取遥感图像的边界特征，可以实现边界检测，且还可以利用边界特征对包含上下文信息的多层级特征进行增强，实现边界对语义上下文的引导，如此，可以使模型具有更细致的上下文视角，有助于提高模型进行语义分割的性能。