【摘要】 针对遥感土地图像中来自异源图像轮廓纹理不明显及差异图特征表达不准确导致变化检测不准确问题，提出了基于梯度信息与差异图融合的变化检测算法。首先，梯度信息图像作为输入源，增强了图像的轮廓及纹理；其次，设计了一种基于区域特征融合与加权平均融合差异图融合方法，通过自编码网络将异源图像映射到同一空间进行差分及像素级融合。其中，自编码器网络的输入使用了形态学梯度、方向梯度的融合图像，实验结果表明性能良好。更多还原

【Abstract】 In order to solve the problem of inaccurate change detection caused by the indistinct contour texture of remote sensing land images and the inaccurate expression of difference map features, a change detection algorithm based on the fusion of gradient information and difference map is proposed. Firstly, the gradient information image is used as the input source to enhance the contour and texture of the image. Secondly, a difference map fusion method based on regional feature fusion and weighted average fusion is designed, which maps the heterogeneous images to the same space through self-coding network for difference and pixel level fusion. Among them, the input of self-encoder network uses morphological gradient and directional gradient fusion images,and the experimental results show that the performance is good.更多还原