An automatic registration method for hyperspectral and lidar images based on clustering and optical flow

Abstract

A method for automatic registration of hyperspectral and LiDar images based on clustering and optical flow methods provided by this application includes: selecting target bands in hyperspectral images, and constructing pseudo-color images based on target bands; pre-processing LiDar images processing to obtain a two-dimensional rasterized image; use the PCA algorithm to reduce the dimensionality of the pseudo-color image and the two-dimensional rasterized image and extract the principal components to obtain the first data set and the second data set respectively; use K-means aggregation The algorithm determines the seed points of the corresponding areas in the first data set and the second data set, and performs rough image registration operation to obtain the initial registration image; uses the optical flow method to perform fine registration on the initial registration image to obtain the registration image. This application combines the PCA algorithm with the K-means clustering algorithm to realize automatic extraction of registration seed points, which improves the registration efficiency, and uses the optical flow method for fine registration to improve the registration accuracy.

Description

technical field [0001] This application relates to the field of remote sensing image processing, in particular to an automatic registration method for hyperspectral and LiDar images based on clustering and optical flow methods. Background technique [0002] With the development of science and technology and social life, in the field of remote sensing image processing and application, the fusion of multi-source data and collaborative processing has become the general trend. Remote sensing image registration is the process of calibrating, superimposing, and matching corresponding points of two or more remote sensing images of objects in the same area and at different times, different conditions, different sensors, and different shooting angles. LiDar images are a set of irregular 3D space point clouds in a unified coordinate system, each point has three-dimensional coordinates X, Y, Z and the intensity value I returned by the laser; LiDar images lack color information, and it ...

AI Technical Summary

Problems solved by technology

[0003] The traditional hyperspectral and LiDar image registration method is to manually select the registration control seed point for registration, resulting in a large additional error, and using the nearest point iteration (ICP), the calculation is very large

Images