Hyperspectral image classification algorithm based on wave band clustering and improved domain transformation recursive filtering

Abstract

The invention aims to provide a hyperspectral image classification algorithm based on wave band clustering and improved recursive filtering, a central wave band in each wave band subset is iteratively found out by utilizing a wave band clustering algorithm based on relative entropy, and information of an original spectral wave band is reserved. Compared with an original domain transformation recursive filtering algorithm, the hyperspectral image classification algorithm is advantageous in that Gaussian filtering is carried out on the central wave band set obtained through clustering to serve as a guide image of domain transformation recursive filtering, meanwhile, an improved domain transformation recursive filtering algorithm is carried out on the central wave band of each set, finally, a feature image set of all the central wave bands is obtained, spatial-spectral joint information of a hyperspectral image is fully obtained, and the subsequent classification precision is improved.

Description

technical field [0001] The invention belongs to the technical field of image processing, and in particular relates to a hyperspectral image classification algorithm based on band clustering and improved domain transformation recursive filtering. Background technique [0002] Hyperspectral remote sensing refers to the technology that uses many narrow electromagnetic wave bands to obtain object-related data. This technology detects and identifies targets through multiple spaces such as spectral space and feature space. In recent years, hyperspectral images have been widely used in aerospace, agriculture, etc. Science, geographical monitoring and environmental protection and other interdisciplinary fields, among which improving the classification accuracy of hyperspectral images is one of the current research hotspots. However, the existing hyperspectral image object recognition and classification algorithms are far from meeting the rapid development and demands of hyperspectra...

AI Technical Summary

Problems solved by technology

Related techniques of feature extraction include principal component analysis, local linear embedding, and linear discriminant information. Although feature extraction technology can achieve higher classification accuracy, the low-dimensional expression extracted by feature extraction has no physical meaning and is difficult to explain.

Band selection technology generates corresponding low-dimensional representations by selecting the most important spectral bands, which can retain the original information of spectral bands, but the improvement of classification accuracy is limited.

Images