基于深度可分离卷积和改进空间金字塔的高分辨率遥感影像变化检测

Change Detection in High Resolution Remote Sensing Imagery Based on Depthwise Separable Convolution and Improved Spatial Feature Pyramid

建筑物变化检测是变化检测领域的一个重要研究方向,目前的众多建筑物变化检测方法大多参数量较大,变化检测精度不高。文章提出了一种结合深度可分离卷积和改进空间金字塔的模型。该方法上采样部分以深度可分离卷积代替传统卷积模块,即以逐通道卷积和逐点卷积的方式代替传统卷积模式,在保持精度的同时有效降低了参数量。增加改进空间金字塔结构以结合多尺度特征信息,增加影像多特征表达。实验结果表明。文章提出的方法可以有效提高建筑物数据集中的变化检测准确度。

Building change detection is an important research direction in the field of change detection. Most of the current building change detection methods have large parameters and low change detection accuracy. This paper proposes a model that combines depthwise separable convolutions and improved spatial pyramids. In the upsampling part of this method, the traditional convolution module is replaced by depthwise separable convolution, that is, the traditional convolution mode is replaced by channel-by-channel convolution and point-by-point convolution, which effectively reduces the amount of parameters while maintaining the accuracy. Add an improved spatial pyramid structure to combine multi-scale feature information and increase the multi-feature representation of images. Experimental results show that. The proposed method can effectively improve the accuracy of change detection in building datasets.