【摘要】 遥感图像的地物轮廓提取在实际生产生活中有重要的意义。论文采用以深度残差网络作为生成器,深度卷积神经网络作为判别器,采用循环一致生成对抗网络的训练方式,训练出一个可以用来提取地物轮廓的生成网络。该训练方法可以不使用严格对齐的训练图对。论文采用了灰度图和梯度强度图作为生成器的输入,对比分析了两种输入方法的效果和性能,验证了网络的泛化能力。更多还原

【Abstract】 Remote sensing image of the contours of the main object in the actual production of life has important significance.In this paper,depth residual network is used as generator and deep convolution neural network is used as discriminator,and a cyclic and consistent generation of training methods are adopted to against network,a generation network is trained that can be used to extract contour of object. This training method may not use strictly aligned training chart pairs. In this paper,the grayscale and gradient intensity maps are used as input of the generator. The effects and performances of the two input methods are compared and the generalization ability of the network is verified.更多还原