一种基于双支路神经网络遥感影像的建筑物变化检测方法

A Remote Sensing Building Change Detection Method Based on Dual-branch Neural Network

针对当前深度学习遥感建筑物变化检测方法难以顾及多时相语义特征的相关性和互补性的问题,本文利用双支神经网络具有从不同时相遥感影像中直接提取图像特征信息,发现不同时相遥感影像中同类地物特征的相似性找出图像中的变化信息的能力,提出了一种基于双支路神经网络遥感影像的建筑物变化检测方法,用于克服不同成像条件带来的图像特征差异,融合多时相语义特征的相关性和互补性,提高遥感建筑物变化检测的精度。在国产高分卫星影像数据集上的实验结果表明,本文方法比单支网络SegNet识别准确性更高,边界更加清晰,小斑块建筑物变化也能够更准确地识别。

In order to solve the problems that the current deep learning method of building change detection cannot effectively take into account the correlation and complementarity of multi-temporal semantic features.In this paper,a remote sensing building change detection method based on dual-branch neural network is proposed to integrate the correlation and complementarity of multi-temporal semantic features,overcome the difference of image features caused by different imaging conditions,and improve the accuracy of building change detection.It uses the characteristics of dual-branch network,which can extract image feature information directly from remote sensing images of different time phases,finds the similarity of similar features in remote sensing images of different time phases and the change information in the image.The experimental results on the GF satellite image datasets show that the proposed method has higher recognition accuracy,clearer boundary,and more accurate recognition of small patch building changes than SegNet network.