A land and water segmentation method for unsupervised domain adaptation of remote sensing images

Abstract

The invention discloses a water and land segmentation method for unsupervised domain adaptation of remote sensing images. Domain adaptation makes the model simple and fast for Landsat satellite remote sensing data. The specific implementation includes steps such as data acquisition, network design, and network training and output. Compared with the prior art, the invention can directly output the recognition result of land and water segmentation for the input Landsat satellite remote sensing data. Compared with the method of field measurement, the invention is simpler, and can directly use the public supervision data set to perform domain adaptation on the neural network. It avoids labeling its own data set, saving a lot of manpower and time.

Description

technical field [0001] The invention relates to the technical field of semantic segmentation of remote sensing images, in particular to an unsupervised domain adaptive land and water segmentation method based on Landsat satellite remote sensing images. Background technique [0002] Semantic segmentation of remote sensing images is the task of assigning a semantic label to each pixel, which is widely used in land use surveys and environmental monitoring. Land and water segmentation plays an important role in coastline extraction and channel mid-dam observation. In recent years, supervised algorithms, especially deep convolutional neural networks (CNNs), have shown impressive performance in semantic segmentation of remote sensing images. Segmentation based on deep learning often requires a large amount of manually labeled data for training, and labeling remote sensing images for each pixel is time-consuming and laborious. [0003] Currently, one approach to this problem is t...

AI Technical Summary

Problems solved by technology

However, there is a discrepancy between the training (source) and test (target) images

Images