Satellite image low-coherence region identification method and device based on deep learning

Abstract

A satellite image low-coherence region identification method and device based on deep learning belong to the field of satellite image processing, and are characterized in that a quality map obtained by SAR image processing is combined with semantic segmentation to form a process learning framework, and a deep convolutional network is introduced for identification. Compared with manual interpretation and a traditional machine learning algorithm, the low-coherence region recognition method based on deep learning greatly reduces dependence on artificial feature engineering and sample data, and a large-range region can be quickly processed based on high efficiency of convolution calculation; meanwhile, on the basis of introducing global attention, relation representation between pixel positions and semantic categories is further described, the recognition effect of inherent high-noise areas of the SAR image is relieved, the recognition precision is improved on the premise that the calculated amount is not remarkably increased, and low-coherence areas are accurately extracted.

Description

technical field [0001] The invention belongs to the field of satellite image processing, in particular to a method and device for identifying low-coherence regions of satellite images based on deep learning. Background technique [0002] Synthetic Aperture Radar Interferometry (InSAR) is a space-to-earth observation technology that uses phase information to extract ground elevation. It has the characteristics of all-day, all-weather, high precision, and large area. It has been widely used and made outstanding contributions to the development of the national economy and national security. [0003] Tianhui-2 satellite is the first microwave and terrain surveying and mapping satellite independently developed by my country based on the formation system. It uses InSAR technology to efficiently obtain high-precision DEM data. However, due to the complexity of InSAR technology, there are still many problems. It has been essentially solved, such as the problem of decoherence of the ...

AI Technical Summary

Problems solved by technology

However, for the low-coherence areas formed by radar shadows, overlaps, and smooth planes (water surfaces) caused by the imaging characteristics of SAR images, plus the time-distance decoherence caused by repeated orbits and formation flight problems, various complex factors Superimposed together, it adds difficulties to the low-coherence area identification of the Tianhui-2 satellite. In addition, the combination of deep learning and SAR image recognition technology still faces challenges in the following aspects: 1. Intra-class differences in SAR low-coherence areas The similarity between classes and classes increases the difficulty of extraction; 2. The SAR images in different seasons and different regions are significantly different, and the prediction ability of the model fluctuates greatly among data sets of different time periods; 3. Because the Tianhui-2 satellite needs to guarantee global surveying and mapping task, the data is massive, but the samples available for supervised training are very limited

Images