Optical remote sensing image saliency target detection method

Abstract

The invention relates to an optical remote sensing image saliency target detection method, and the method comprises the steps: S1, building a dense attention flow network which comprises an attentionflow guided feature coding module and a progressive feature decoding module; s2, importing an optical remote sensing image; s3, processing the optical remote sensing image by adopting an attention flow guided feature coding module to generate enhanced features with higher discrimination capability, the attention flow guided feature coding module mainly comprising a global context perception attention module and a dense attention flow structure, and S4, decoding the enhanced features in the step S3 by adopting a progressive feature decoding module, gradually fusing the deep features and the shallow features in a feature decoding stage, and generating a plurality of side outputs and final outputs under the supervision of the saliency map and the saliency edge map.

Description

Technical field[0001]The invention belongs to the field of remote sensing image processing and deep learning, and relates to a method for detecting a significant target in an optical remote sensing image.Background technique[0002]Inspired by the human visual attention mechanism, the purpose of the visual saliency detection task is to detect the most concerned target or area in the input data (such as images, videos, etc.). It has been widely used in target detection, image editing, intelligent photography, and autonomous driving Many other fields have important research value and broad market prospects. However, the optical remote sensing image that the present invention focuses on has some properties that are different from the images taken by traditional handheld cameras (also called natural scene images) due to its special shooting method and imaging environment, which makes the direct transplantation of existing natural scene images significant Target detection methods often fai...

AI Technical Summary

Problems solved by technology

[0004] (1) The existing deep learning optical remote sensing saliency target detection method usually directly propagates and fuses multi-level features, and does not consider the interactive relationship between the attention information corresponding to each level of features;

[0005] (2) Since the distribution range of salient targets in optical remote sensing images may be relatively large or there may be multiple similar targets, existing methods cannot detect them completely, resulting in low integrity of salient detection results

Images