Remote sensing image cloud detection method based on feature multi-scale perception and adaptive aggregation

Abstract

The invention provides a remote sensing image cloud detection method based on feature multi-scale perception and adaptive aggregation, and the method comprises the steps: a feature multi-scale perception module perceives context information of different scales through parallel expansion convolution pairs between an encoder and a decoder because the semantic information of clouds and cloud shadows on different scales is different; the feature adaptive aggregation module adaptively learns contributions of features of different scales to cloud and cloud shadow detection based on a self-attention mechanism, and realizes weighted aggregation of multi-scale features; the vector post-processing flow comprises operations such as hole filling, buffer area analysis, edge smoothing and erasing, so that the problems that a part of areas of a deep learning network detection result are fine and are zigzag are solved, an image effective area closer to manual production is finally obtained, and even the image effective area is more accurate than the image effective area obtained by manual production; a large amount of time and labor cost are saved for remote sensing image production, the finally obtained detection effect is good, and subsequent application of the image is facilitated.

Description

【Technical field】 [0001] The invention belongs to the technical field of remote sensing image cloud and cloud shadow automatic segmentation, relates to a remote sensing impact cloud detection method, in particular to a remote sensing image cloud detection method based on feature multi-scale perception and adaptive aggregation, which uses semantic segmentation network and Vector post-processing adaptive extraction of clouds and cloud shadows in remote sensing images. 【Background technique】 [0002] In recent years, my country's earth observation technology has developed vigorously, satellite images have increased rapidly, and the degree of marketization has increased year by year. At present, the related applications of optical remote sensing satellite images still play a major role in earth observation. In optical images, clouds, as a widespread object, are an important factor in the fifth analysis of images and the accuracy of remote sensing image feature extraction. The ...

AI Technical Summary

Problems solved by technology

Because of its excellent performance in tasks such as feature extraction and classification, it is widely used in many fields. In the field of cloud detection, many studies combine classical machine learning algorithms with texture analysis methods, and combine classical machine learning algorithms with spatial texture analysis methods. It can better classify the input features, and is more universal than the traditional rule-based cloud detection method, but it still needs to manually design features, and the detection effect on cloud shadows is poor

Images