Remote sensing image typical surface feature extraction method based on multi-task attention mechanism

Abstract

The invention discloses a remote sensing image typical surface feature extraction method based on a multi-task attention mechanism, and the method employs four attention modules to fuse global features from the interior and the exterior, enlarges the receptive field of a model, and solves the problems of wide surface feature element distribution range and large region area. A multi-decoder structure is constructed by using a multi-task mechanism, so that competition of different ground feature types for model parameters is reduced, and misjudgment of similar ground features is reduced; the edge constraint is increased by utilizing the edge extraction task and the distance map extraction task, so that the edge extraction effect is improved, intelligent extraction of multiple types of typical surface feature elements is finally realized, and the extraction precision is relatively high.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing, and more particularly relates to a method for extracting typical features of remote sensing images based on a multi-task attention mechanism. Background technique [0002] As the spatial resolution of remote sensing images is getting higher and higher, the structure, shape, texture and other details of the ground objects on the remote sensing images become clear and distinguishable to the naked eye, which makes the use of remote sensing images to obtain the spatial information of typical ground objects. distribution becomes possible. The extraction of typical features is one of the core technologies of the high-resolution remote sensing application service chain, and has played a huge role in many fields such as global change, disaster detection, and resource management and control. In the field of global change, the spatial distribution of buildings, forests, grasslands,...

AI Technical Summary

Problems solved by technology

Before 2015, deep learning technology was only widely used in image classification and detection, but in the field of image segmentation, due to the high time complexity of dense prediction tasks, it is difficult to get real applications

However, the classic UNet and other models have a limited receptive field, and the accuracy is usually low when extracting large-scale ground objects in remote sensing images; in addition, some ground objects in remote sensing images have similar characteristics, which need to be comprehensively judged based on surrounding context information. This poses a challenge to the global feature acquisition ability of the deep learning model; in addition, the edge extraction effect of the deep learning model on different object categories is usually poor, and the boundaries are usually not clear enough

Images