Remote sensing image registration method and system fusing SIFT feature and CNN feature

Abstract

The invention provides a remote sensing image registration method and system fusing SIFT features and CNN features, which comprises extracting feature points of an input reference image and an image to be registered, using SIFT mode to carry out feature expression on a neighborhood area of the feature points, and obtaining SIFT features; the neighborhood region of feature points being used as theinput of convolution neural network (CNN), and the CNN model trained by transfer learning strategy being used to express the high-level features, and the CNN features being obtained. The SIFT featureand CNN feature are fused, similarity is calculated, geometrical transformation parameters between reference image and the image to be registered are estimated, geometrical transformation and resampling are performed on the image to be registered, and the registered image is obtained. The invention combines the bottom feature extracted by the traditional feature and the advanced feature extractedby the convolution neural network, and the combined feature can more accurately express the remote sensing image content, greatly improves the remote sensing image registration accuracy and has strongadaptability.

Description

technical field [0001] The invention belongs to the technical field of remote sensing image processing, and relates to a remote sensing image registration method and system that integrates SIFT features and CNN features. Background technique [0002] Image registration is the process of geometrically correcting unregistered images based on georeferenced images that contain the same area, possibly from different shooting times, different sensors, or different shooting angles. Image registration is a basic problem in the field of remote sensing image processing, which has a significant impact on subsequent applications, such as image fusion and change detection. [0003] Image registration is mainly divided into three steps: image matching, geometric transformation parameter estimation and image transformation. Image matching is the basis of image registration. Therefore, similar to image matching, image registration methods can be roughly divided into region-based matching m...

AI Technical Summary

Problems solved by technology

However, the feature extraction method based on deep learning relies heavily on samples. At present, there are many sample libraries for natural images, but there are almost no sample libraries for remote sensing images. It is expensive to use manual annotation of remote sensing images.

Moreover, at present, feature extraction methods based on deep learning have been applied in the fields of remote sensing image retrieval, classification and target recognition, but these methods are not specially designed for remote sensing image registration, and lack corresponding inspiration.

Images