Remote sensing image semantic segmentation method based on self-supervised contrast learning

Abstract

The invention discloses a remote sensing image semantic segmentation method based on self-supervised contrast learning. The method comprises the following steps: constructing a semantic segmentation network model (such as Deeplab v3 +); carrying out pre-training on an encoder of the network model by adopting unlabeled data; after the pre-training is completed, performing supervised semantic segmentation training on the network model on a labeled sample; carrying out semantic segmentation on the remote sensing image by adopting the network model subjected to supervised semantic segmentation training, wherein in the pre-training process, a mode of combining global style comparison and local matching comparison is adopted to carry out comparative learning. Compared with the prior art, comparison self-supervised learning is applied to a remote sensing semantic segmentation data set, a global style and local matching comparison learning framework is provided, and the remote sensing image semantic segmentation method based on self-supervised comparison learning is formed, so that the semantic segmentation method is wider in application range and better in segmentation effect.

Description

Technical field [0001] The present invention relates to the field of remote sensing image semantics, and more particularly to remote sensing image semantic segmentation methods based on self-supervision comparison learning. Background technique [0002] With the development of remote sensing technology, the acquisition of high-score-sensing images is increasingly easy, remote sensing images have become more and more extensive applications in urban planning, disaster monitoring, environmental protection, and transportation. Extraction and identification of information in remote sensing images is usually the basis of all applications, and semantic segmentation is a technique for identifying classification of full-image, so it has always been important and challenging in the field of remote sensing. [0003] In recent years, the development of deep learning technology has achieved impressive results in semantic segmentation, and is increasingly widely used in global surface coverage...

AI Technical Summary

Problems solved by technology

[0008] In view of this, the purpose of the present invention is to solve the problem of directly learning features from unlabeled images to help downstream semantic segmentation tasks with only a small amount of annotations. In this paper, the contrastive self-supervised learning is applied to the remote sensing semantic segmentation data set, a global style and local matching contrastive learning framework is proposed, and a remote sensing image semantic segmentation method based on self-supervised contrastive learning is formed.

Images