Multi-temporal hyper-spectral image classification method based on spatial-spectral feature preserving global geometric structure

Abstract

Disclosed is a multi-temporal hyper-spectral image classification method based on a spatial-spectral feature preserving global geometric structure. The invention relates to a multi-temporal hyper-spectral remote sensing image classification method. The invention aims to solve the problem that a hyper-spectral multi-temporal data tag is hard to acquire, and classification of data of a target time phase by direct use of hyper-spectral data of a source time phase is unreliable under the condition of obvious spectral drift of an image. The method specifically comprises the following steps: (1) inputting X<s> and X<t> and the spatial coordinates Z1, Z2 thereof, and a tag vector Y of corresponding category in each line of X<s>; (2) calculating the spatial-spectral distance of X<s> and X<t>, and selecting nearest points as a data pair needing matching; (3) calculating D<s, s>, D<t, t>, and D<s, t>, adjusting the scale of the data set, and building a distance matrix D; (4) getting the mapping matrixes alpha and beta of X<s> and X-wavy line<t> in an alignment space, and getting projections f<s> and f<t>; and (5) performing classification using a KNN classification model according to f<s> and f<t> as well as a tag Y corresponding to f<s> to get a classification tag of the target time phase. The method is used in the field of image classification.

Description

technical field [0001] The invention relates to a multitemporal hyperspectral image classification method. Background technique [0002] With the development of optical sensors and spectroscopic technology, hyperspectral remote sensing imaging technology has developed rapidly on the basis of multispectral remote sensing imaging technology. Hyperspectral imaging technology combines imaging technology and spectral technology, and can simultaneously record the spatial structure information and rich spectral information of the area to be detected, and with the development of remote sensing technology, it can now provide large-area areas with spatial and temporal sequence Hyperspectral multi-temporal image data for ground object perception and monitoring. Multi-temporal hyperspectral images make multi-temporal analysis, multi-angle research, and accurate ground object change detection possible. However, it is unrealistic to provide sufficient label information for each phase ma...

AI Technical Summary

Problems solved by technology

[0003] Multi-temporal hyperspectral data classification mainly faces three main problems: 1. The number of bands continues to increase, resulting in information redundancy and increased data processing complexity; 2. It is very difficult to obtain labeled samples of hyperspectral images and requires a large amount of Manpower, material resources and time-consuming; 3. There may be spectral drift in multi-temporal hyperspectral images, resulting in unreliable spectral similarity of image data from different temporal phases

These are all methods of maintaining the local geometric structure of the data, but this method is sometimes not applicable, for example, for highly folded data, it may bring wrong spectral neighbors and affect the classification effect

Images