Remote sensing image space-time fusion method based on hybrid convolutional network

Abstract

The invention discloses a hybrid convolutional network-based remote sensing image space-time fusion method, which comprises the following steps of: inputting three multi-band remote sensing images into a network model, and splitting the three multi-band images into single-band images in the model; then, according to the characteristics of a remote sensing image space-time fusion task, sending the single-band image into a double-branch network according to a certain band combination and arrangement sequence for feature extraction, and completing single-band feature extraction and single-band image reconstruction; the 2D-CNN branch is mainly used for extracting spatial information features, the 3D-CNN has one more dimension than the 2D-CNN, the dimension is in a time-space fusion task, namely, the time dimension, and therefore time change features and spatial detail features are extracted at the same time through the 3D-CNN branch. The method can generate the high-temporal-spatial-resolution remote sensing image.

Description

technical field [0001] The invention belongs to the field of improving the temporal resolution and spatial resolution of remote sensing images. Specifically, it relates to a spatiotemporal fusion method based on a hybrid convolutional network, which uses a dual-branch network feature extraction and fusion, and adopts a band iterative method to generate remote sensing images with both high temporal resolution and high spatial resolution. Background technique [0002] The rapid development of remote sensing earth observation technology provides new methods and means for human beings to understand the world objectively and comprehensively. As scientific research on the application of remote sensing images becomes more and more extensive and in-depth, human beings can obtain electromagnetic radiation data reflected or emitted by various landscapes such as land, atmosphere, and ocean through various remote sensing observation instruments on aerospace platforms. After humans have...

AI Technical Summary

Problems solved by technology

[0013] The STARFM model can accurately invert phenological changes, but its performance depends on the characteristic patch size of the land surface landscape, and the performance will decrease when inverting a land landscape with extremely uneven distribution

The algorithm developed based on FSDAF theory still has two main problems: (1) Due to the input of low-resolution pixels containing type change information and a large amount of boundary information for decomposition calculation, FSDAF is prone to lose spatial details and predict "more blurred" images

(2) FSDAF cannot accurately retrieve land cover change areas

To sum up, the challenges that the spatio-temporal fusion method still faces are: 1) Accurate inversion of land cover change, especially for areas of land cover type change, accurate inversion is conducive to monitoring the dynamics of the land surface and the earth system

Images