Deep learning change detection method based on radar remote sensing data

Abstract

The invention belongs to the technical field of land coverage change detection, and particularly relates to a deep learning change detection method based on radar remote sensing data. According to the invention, the defect that optical remote sensing data are difficult to acquire in cloudy and rainy areas is avoided to a certain extent by using the complete polarization radar remote sensing image, and false changes caused by seasonal factors are avoided to a certain extent by using data of multiple months of the previous and later years when the data set is created; multi-dimensional initial features are extracted through front and rear time phase multi-scene complete polarization radar images, and thus constructing a difference image according to the initial feature image; meanwhile, aiming at the small sample problem, a depth separable convolution structure of a MobileNet lightweight network model is adopted, the structure is organically combined with a U-Net basic framework, and the constructed lightweight model is used for change detection, so that a change detection result is effectively obtained on a small sample data set. Finally, the change detection result is automatically, efficiently and accurately extracted.

Description

technical field [0001] The invention belongs to the technical field of land cover change detection, and in particular relates to a deep learning change detection method based on radar remote sensing data. Background technique [0002] Remote sensing image change detection uses multi-temporal remote sensing images and uses change detection algorithms to extract change information between different temporal remote sensing images. This technology has been maturely applied to land cover / land use, natural disaster assessment, and urbanization construction. and crop growth. [0003] Scholars at home and abroad have carried out some research on change detection using optical remote sensing images. Shen et al. proposed a classification framework for the impact of non-adjacent pixels on classification, using a fully convolutional network (Fully Convolution Network, FCN) to incorporate remote context information. Good classification results are obtained on hyperspectral datasets. Ly...

AI Technical Summary

Problems solved by technology

[0005] In summary, although methods based on deep learning are gradually applied to the change detection task of SAR images, due to the high cost of SAR image acquisition and the difficulty of visual interpretation due to the special imaging mechanism, the change detection There are few relevant public data sets, so the research on change detection using deep learning methods is not sufficient, and most public data sets only contain dual-polarization or single-polarization information but not full-polarization information, which greatly hinders The development of multi-polarization SAR in the field of change detection

Images