关键词: 深度学习, 变化检测, U-Net, ResNet, 孪生神经网络, 遥感监测, 自然资源监测

Abstract: Intelligent extraction of change detection from remote sensing images is the basis of dynamic monitoring of natural resources. This paper briefly describes the history and characteristics of remote sensing image change detection technology, proposes to use three deep learning algorithms ResNet, U-Net and siamese neural network at the same time, and designs an intelligent extraction platform of remote sensing image change detection that integrates “image preprocessing, intelligent detection, and collaborative screening”, and describes the design ideas of each module in detail. Practice has shown that the integration of three deep learning algorithms is beneficial to solve problems such as difficult transformation and limited scope of application of a single deep learning algorithm, effectively improving the recall rate of remote sensing image change detection, and improving work efficiency compared with visual interpretation more than 3 times. The research results have been widely used in the “1+N” satellite monitoring of natural resources in Hunan Province.

Key words: deep learning, change detection, U-Net, ResNet, siamese neural network, remote sensing monitoring, natural resource monitoring