Kernel joint sparse representation hyperspectral image classification method based on dictionary optimization

Abstract

The invention relates to the technical field of remote sensing image processing, in particular to a dictionary optimization-based kernel joint sparse representation hyperspectral image classification method, which comprises the following steps of: extracting a principal component feature map of a hyperspectral image and performing reconstruction-based morphological processing; performing super-pixel segmentation on the image after morphological processing; lBP texture features are extracted, and pixels of feature positions in the middle of each block of superpixels are used as dictionary alternative subsets; in each class of dictionary alternative subsets, each pixel is used as a dictionary, residual errors of other pixels in the class of dictionary alternative subsets are calculated, average values are calculated, the average values are arranged in an ascending order, and the pixels corresponding to the first 5% average values form an optimized dictionary subset; obtaining a classification result of the image by using a kernel joint sparse representation method under double constraints of a superpixel edge and a fixed neighborhood; the method can effectively improve the image classification precision.

Description

technical field [0001] The invention relates to the technical field of remote sensing image processing, in particular to a hyperspectral image classification method based on dictionary optimization based on kernel joint sparse representation. Background technique [0002] Hyperspectral imagery (HSI) is a high-dimensional image with a huge amount of data, which can reflect the spatial information and spectral information of ground objects. It is widely used in remote sensing, environmental monitoring, urban mapping and target recognition and other fields. In recent decades, HSI classification has been a popular research topic in the field of remote sensing. The core idea is to use spectral information and extracted meaningful spatial information to classify each spectral pixel. To achieve this goal, many methods have been developed. Support Vector Machine (SVM), Multinomial Logistic Regression (MLR), Neural Network, Artificial Immune Network are all widely used classifiers, ...

AI Technical Summary

Problems solved by technology

In the traditional sparse representation classification algorithm, atoms are randomly selected in each type of sample to form a dictionary set. Although this selection method has low algorithm complexity, it cannot guarantee the sparse reconstruction of the pixels to be tested, which affects the quality of the image. classification accuracy

Images