Interactive classification method for hyperspectral images based on kernel cooperative representation

Abstract

The invention discloses an interactive classification method of hyperspectral images based on nuclear cooperative representation, which comprises the following steps: loading or manually selecting a training area of interest for hyperspectral image data; taking the pixels in the training area as the training set, the hyperspectral image probability distribution map being obtained by using the kernel cooperative representation to classify the images; taking the pixels in the hyperspectral image as nodes, the spatial-spectral joint map being established according to the spatial positional relationship of the pixels. The probability distribution map and the joint space-spectrum map are put into the post-processing model to predict the classification results. According to the classification prediction results, part of the training areas are added or deleted repeatedly manually, and the refined classification results are predicted. At first, the invention carry out joint classification of coarse-grained core cooperative representation space spectrum, and then through real-time addition and deletion of part of training area, the invention carry out joint classification of fine-grained core cooperative representation space spectrum, so that the classification mapping accuracy and efficiency can be remarkably improved.

Description

technical field [0001] The invention relates to the technical field of hyperspectral image processing, in particular to an interactive hyperspectral image classification method based on kernel cooperative representation. Background technique [0002] Hyperspectral remote sensing sensors can acquire hundreds of images in many continuous and narrow wavelength bands, and their imaging areas cover the visible light to infrared regions. The high spectral resolution of hyperspectral remote sensing images makes them widely used in the fields of surface target detection, urban planning, agricultural early warning, and military reconnaissance. [0003] Hyperspectral remote sensing image classification is an important link in the application of hyperspectral remote sensing images. At present, the more commonly used classification methods are: support vector machine, multinomial logistic regression, sparse representation and collaborative representation, etc. Because these classifica...

AI Technical Summary

Problems solved by technology

[0004] For supervised hyperspectral remote sensing image classification, the acquisition of data labels is an extremely time-consuming and laborious work

Moreover, due to the variability of imaging conditions, the universality of labels on different datasets is often poor

Images