Deep network multi-source spectral image fusion method for multi-supervised recursive learning

Abstract

The invention discloses a deep network multi-source spectral image fusion method for multi-supervised recursive learning, and the method comprises the steps: employing recursive learning to form recursive residual sub-networks, and adding the output and input of each recursive residual sub-network to serve as the input of a next recursive residual sub-network; wherein the network is composed of a pre-super-resolution module and a fusion module, the pre-super-resolution module realizes automatic learning of up-sampling interpolation, and a pre-super-resolution image and a multispectral image are spliced as input of the fusion module; establishing a pre-super-resolution module and a fusion module by adopting a plurality of recursive residual sub-network stacking methods; adopting a multi-supervised learning mode, and splicing low-level, intermediate-level and high-level features and convolutional layers to form intermediate fusion images of all levels; and taking the L1 norm and the spectral angle as two metrics of a loss function, establishing a joint loss function between the intermediate fusion image of each level and the real image, and carrying out end-to-end network training. Simulation experiment results prove the effectiveness of the method for multi-far-spectrum image fusion.

Description

technical field [0001] The invention relates to the field of hyperspectral image fusion, in particular to a multi-supervised recursive learning deep network multi-source spectral image fusion method. Background technique [0002] In recent years, deep learning has become a research hotspot in the field of artificial intelligence, has received extensive attention from the theoretical and industrial circles, and has been widely used in pattern recognition, computer vision, natural language processing and other fields. The deep learning model is generally a neural network with a multi-layer structure. Multi-level feature extraction is performed on the data through multiple nonlinear transformations of the multi-layer neural network, and the hierarchical features from low-level to high-level are automatically learned. As the number of layers increases, the feature The degree of abstraction also increases. Compared with the traditional shallow machine learning model, the feature...

AI Technical Summary

Problems solved by technology

[0004] However, most of the above models are shallow in depth and cannot fully utilize the powerful feature extraction and nonlinear representation capabilities of deep network structures.

Images