Self-supervised learning fusion method for multi-band images

Abstract

The invention relates to a fusion method for multi-band images, in particular to a self-supervised learning fusion method for multi-band images based on a multi-discriminator generative adversarial network, which comprises the following steps of: designing and constructing a generative adversarial network which consists of a generator and a plurality of discriminators, and a label image is a multi-band source image; wherein a generator network structure is composed of a feature enhancement module and a feature fusion module which are conceived, acquiring a generation model through dynamic balance training of a generator and a discriminator to acquire a multi-band image fusion result. A multi-band image end-to-end self-supervision fusion neural network is realized, a result has better definition and information amount, the detail information is richer, and the method better conforms to the visual characteristics of human eyes.

Description

technical field [0001] The invention relates to an image fusion method, in particular to a multi-band image fusion method, specifically a self-supervised learning fusion method for multi-band images. Background technique [0002] At present, high-precision detection systems have generally adopted wide-spectrum multi-band imaging, and existing researches are mainly carried out on two bands of infrared and visible light. Therefore, it is imminent to explore the synchronous fusion of multiple (≥3) images. In recent years, image fusion research based on deep artificial neural networks has emerged. However, due to the lack of standard fusion results in the field of image fusion, that is, there is generally a lack of label data when using deep learning to build image fusion models, resulting in difficulties in deep learning training or poor fusion effects. The more the number of synchronously fused images, the more prominent the problem will be. [0003] Self-supervised learning ...

AI Technical Summary

Problems solved by technology

[0004] In order to solve the problem of limited fusion results due to the lack of label images when using deep learning methods in the field of image fusion to fuse multi-band images, the present invention proposes a new method for self-supervised learning and fusion of multi-band images based on multi-discriminator generation confrontation network

Images