Hyper-spectral image classification method based on hyper-pixel segmentation and two-stage classification strategy

Abstract

The invention provides a hyper-spectral image classification method based on hyper-pixel segmentation and a two-stage classification strategy. The hyper-spectral image classification method comprisesthe following steps: A, preparing a hyper-spectral image to be processed and an initial training sample data set; B, performing super-pixel segmentation processing on the hyper-spectral image, judgingwhether each piece of super-pixel data in the hyper-spectral image contains initial training sample data; if so, when the initial training sample data contained in the super-pixel data only belong toone class, classifying all the data in the super-pixel data into the same class as the initial training sample data, and adding the classified super-pixel data into an initial training sample data set to generate an expanded training sample data set; and C, judging whether the data in the hyper-spectral image is classified into one class, and if not, performing second classification processing onthe data which are not classified based on the expanded training sample data set.

Description

Technical field [0001] The invention relates to the field of hyperspectral image processing, in particular to a hyperspectral image classification method based on superpixel segmentation and a two-stage classification strategy. Background technique [0002] Spectral-space hyperspectral image classification (HIC) method is an advanced HIC method. Compared with those methods that only use spectral information, the spectral-space HIC method combines spectral and spatial information to further improve the classification accuracy. In recent years, methods based on superpixels have received more and more attention. Superpixels consist of many pixels in a small area with similar spectral characteristics. Superpixels play an important role in characterizing spatial background information in HSI. [0003] Due to the powerful expressing ability of superpixels on the spatial structure of objects in HSI, many spectral-space HIC methods based on superpixel segmentation have been proposed in ...

AI Technical Summary

Problems solved by technology

[0006] Aiming at the small label training sample size problem in spectral-spatial hyperspectral image classification, the present invention proposes a hyperspectral image classification method based on superpixel segmentation and two-stage classification strategy

Images