Remote sensing image matching method based on deep learning and multi-subgraph matching

Abstract

The invention relates to the field of remote sensing image processing, and provides a remote sensing image matching method based on deep learning and multi-subgraph matching. The method comprises: providing a neural network for generating subgraphs of a remote sensing image; inputting a to-be-matched remote sensing image and a reference image into the neural network to generate a sub-graph; performing feature extraction on the sub-graphs to obtain a distribution graph of feature points of the sub-graphs; performing feature matching on the feature points to select an optimal feature pair set; and taking a union set from the optimal feature pair set, and mapping the union set back to the remote sensing image to be matched. According to the method, integral image matching of remote sensing images in the prior art is replaced by image matching of multiple sub-images, and the remote sensing image sub-images are generated in a deep learning mode, so that effective semantic information of the remote sensing images can be extracted, information points with unknown semantics in the remote sensing images are filtered out, redundant information of the remote sensing images is filtered out, the remote sensing image quality is improved, and the remote sensing image matching precision and efficiency are improved.

Description

technical field [0001] The present invention generally relates to the field of remote sensing image processing, in particular to a remote sensing image matching method based on deep learning and multi-subgraph matching. Background technique [0002] The task of image matching is to find the correspondence between pixels in two or more images. At present, there are mainly two types of matching methods for remote sensing images, one is the matching method based on the region, and the other is the matching method based on the feature. Among the two types of methods, the feature-based matching method has a small amount of calculation, good robustness, and is not sensitive to image deformation. Since the imaging of remote sensing images will be affected by factors such as noise, seasonal changes, and illumination changes, the feature-based matching method is a more appropriate and commonly used method. [0003] Feature-based image matching methods mainly include three steps: fe...

AI Technical Summary

Problems solved by technology

[0007] For remote sensing images, since remote sensing images have the characteristics of rich texture details and repeated structures, they are easily affected by the redundant information of the image background when performing feature point matching, resulting in a high error matching rate of feature points and low matching performance.

[0008] Specifically, the remote sensing image matching methods in the prior art have the following problems: when performing feature extraction on remote sensing images, it is easy to extract a large number of very similar feature points; in the process of feature matching, due to the global Feature point matching will be easily affected by the background information of the image, resulting in a high error matching rate of feature points; and due to the complexity of the image information, the performance of feature matching will be low

Images