Multispectral image fusion method based on interactive feature embedding

Abstract

The invention provides a multispectral image fusion method based on interactive feature embedding, and belongs to the field of computer vision. The method comprises the following steps: collecting multispectral image pairs, preprocessing the image pairs, including height and width adjustment and sliding window image pair acquisition, and acquiring network training data set; designing a multispectral image fusion network based on interactive feature embedding of self-supervised learning; designing a loss function, and supervising network model training; and in the testing process, inputting a multispectral image pair, and outputting a final image fusion result through network. The network feature extraction capability can be effectively improved, and important information in a fusion result can be reserved.

Description

technical field [0001] The invention belongs to the field of computer vision, in particular to a multispectral image fusion based on interactive feature embedding. Background technique [0002] Multispectral image fusion is to integrate the image features of the same scene captured by multispectral detectors to describe the scene information more comprehensively and accurately. Multispectral image fusion is a part of image fusion tasks and has a wide range of applications in many aspects, such as scene monitoring [1], target recognition, geological survey and military aspects. [0003] Deep learning techniques play an important role in image fusion. Existing image fusion methods based on deep learning are mainly divided into two categories: fusion methods based on adversarial networks and fusion methods based on non-adversarial networks. The fusion method based on the adversarial network aims to fuse the main features of the source image by designing a loss function during...

AI Technical Summary

Problems solved by technology

However, this type of method has the following limitations: it is difficult to optimize the network and it is difficult to design a loss function that contains all important information of the source image

In fusion methods based on non-adversarial networks, the feature extraction process is often implemented in an unsupervised manner, and it is difficult to ensure that the feature extraction

Therefore, whether it is adversarial learning based on loss function design or unsupervised learning, ignoring any important information in the source image (such as gradient, edge, texture, intensity, and contrast) will lead to the loss of important features in the fusion result.

Images