Hyperspectral image preprocessing method for unmixing of abnormal small targets

Abstract

The invention discloses a hyperspectral image preprocessing method for unmixing of abnormal small targets so that a defect that abnormal small targets are easy to ignore according to the existing hyperspectral image preprocessing method serving the unmixing task can be solved. On the basis of characteristic of the small spatial scale of the target, a suspected target on a spatial dimension is determined by using a sliding window, a similarity measure weight is set based on a neighboring pixel position, and influences on determination of specificity of a to-be-measured pixel space by differentneighboring pixels in the window are treated in different ways; on the basis of the characteristic of specificity of a target spectrum relative to a background spectrum, suspected target determinationis carried at the feature dimension by using PCA transform; and then hyperspectral data are screened by combining a K-means method with an orthogonal subspace projection (OSP) method, so that the to-be-processed data volume is reduced effectively and the unmixing precision is improved. The great improvement space exists in engineering application. According to the invention, the follow-up end element extraction stage does not need to be modified; and the algorithm application is flexible.

Description

technical field [0001] The present invention relates to the technical field of remote sensing image processing, in particular to a hyperspectral image preprocessing method for combining spatial spectral information and unmixing tasks of objects with abnormal spectral characteristics and small spatial scales (that is, abnormally small objects). Background technique [0002] Due to its high spectral resolution and wide band range, hyperspectral images have been widely used in military and civilian fields. However, due to the limitation of its spatial resolution, the spectra of different ground objects in the collection scene are mixed with each other, resulting in mixed pixels. Mixed pixel decomposition is a unique and important research content in hyperspectral remote sensing technology, and it is of great significance in fine processing and quantitative analysis of hyperspectral images. In order to improve the unmixing accuracy, unmixing algorithms based on different charac...

AI Technical Summary

Problems solved by technology

[0005] In view of this, the present invention provides a hyperspectral image preprocessing method for unmixing of abnormally small targets, which can solve the disadvantage that the hyperspectral image preprocessing method serving the unmixing task tends to ignore abnormally small targets

Images