Infrared remote sensing water classifier construction method based on transfer learning

Abstract

The invention discloses an infrared remote sensing water classifier construction method based on transfer learning. The method comprises the following steps: firstly, constructing a visible light remote sensing image (source domain) sample set with a label and an unlabeled infrared remote sensing image (target domain) sample set; secondly, extracting a local binary pattern feature of the visible light source domain sample, and training a support vector machine classifier in combination with a tag to initially classify the infrared target domain sample to obtain a pseudo tag of the infrared sample for simulating probability distribution of the target domain sample; based on a transfer learning theory, mapping the feature data of the source domain and the target domain into the same subspaceby taking minimization of a distribution difference distance between the source domain and the target domain as a target, and continuously optimizing a feature mapping matrix by correcting a pseudo label of a target domain sample; and finally, in the subspace, utilizing the optimized feature mapping matrix to transform the source domain feature data, and training a final infrared remote sensing water classifier based on the transformed source domain feature data and tags thereof.

Description

technical field [0001] The invention relates to a method for constructing an infrared remote sensing water body classifier based on migration learning, which belongs to the field of image processing. Background technique [0002] Water resources are an essential resource for human survival. Real-time and accurate access to water body information is of great significance to the effective management and rational use of water resources. With the development of satellite remote sensing technology, a large number of visible light, infrared and other remote sensing water body images can be obtained through sensors. However, most infrared remote sensing water body images are not labeled. [0003] Some infrared image recognition and classification methods for non-water bodies are disclosed in the prior art. For example, the Chinese Invention Patent No. 201610479981.1 discloses a method for detecting dim and small infrared targets using sample feature learning and classification. ...

AI Technical Summary

Problems solved by technology

[0005] However, in this way to construct an infrared remote sensing image water body classification identifier, the main difficulties are: (1) the number of labeled infrared remote sensing image water body samples is scarce, and it is impossible to learn an effective classifier; (2) how to reduce the visible light image The difference in feature distribution between spatial samples and infrared image spatial samples makes the classifier trained by visible light image samples effectively applied in the field of infrared images

Images