基于卷积神经网络的高分辨率遥感影像目标边界提取方法OA

In view of the impact of factors such as occlusion and rotation on object boundary extraction in high-resolution remote sensing images,a CNN-based method for extracting object boundaries from high-resolution remote sensing images is proposed.The high-resolution remote sensing image object boundary extraction framework is implemented by CNN,on the basis of which,a feature enhancement module is introduced to avoid insufficient representation of semantic information and loss of detail information in the shallow and deep feature extraction of network object boundaries.The network loss function is optimized,and the object boundary map is preprocessed and then converted into a probability map of boundary information,and then a threshold value is set to exclude uncertain pixels,so as to enhance the robustness and accuracy of model object boundary extraction.The experimental results show that the proposed method can achieve accurate extraction of object boundaries,and is not easily affected by remote sensing image rotation.In addition,it has excellent ability of object boundary extraction under different degrees of occlusion.