Panchromatic sharpening method for multispectral images based on two-pass convolutional network and hierarchical clstm

Abstract

The present invention is divided into two parts: model training and multi-spectral image panchromatic sharpening. In the model training stage, the original clear multi-spectral and panchromatic images are first down-sampled to obtain simulated training image pairs; And fuse the features of panchromatic images and multispectral images, and combine the hierarchical CLSTM to realize the fusion between multi-level and different depth convolution features; then use the deconvolution network to reconstruct high spatial resolution multispectral images from the fused features image; finally, the Adam algorithm is used to adjust the parameters of the model; in the panchromatic sharpening stage of the multispectral image, firstly, the trained dual-channel convolutional network and hierarchical CLSTM are used to extract and fuse the features of panchromatic and multispectral images. The convolutional network is responsible for extracting the features of multispectral images and panchromatic images and fusing the features selected by itself and CLSTM. CLSTM selects and memorizes different depth features of different depths at multiple levels, thereby realizing the fusion of multi-level and different depth features.

Description

technical field [0001] The invention belongs to the field of remote sensing image processing, and in particular relates to a multi-spectral image panchromatic sharpening method based on a dual-path convolutional network and a multi-level fusion strategy. Background technique [0002] Remote sensing images have two important properties - spectral resolution and spatial resolution. Spectral resolution refers to the minimum wavelength range that the sensor can distinguish when receiving the spectrum of the target radiation. The narrower the wavelength range, the higher the spectral resolution, and the stronger the ability of the sensor to distinguish and identify each band of light in the spectrum, resulting in The more the number of bands, the richer the spectral information of the obtained remote sensing images. Spatial resolution refers to the minimum distance between two adjacent features that can be identified on remote sensing images. The smaller the minimum distance, th...

AI Technical Summary

Problems solved by technology

[0007] In order to overcome the spatial and spectral distortion problems caused by the single convolutional network type, fusion feature level and feature fusion depth in the existing multispectral and panchromatic image fusion methods, the present invention proposes a dual-channel convolutional network and Multi-spectral image panchromatic sharpening method based on multi-level fusion strategy

Images