Non-negative matrix unmixing method based on space-spectrum combined multi-constraint optimization

Abstract

The invention discloses a non-negative matrix unmixing method of air-spectrum combined multi-constraint optimization, which comprises the following steps: 1) estimating the number of hyperspectral terminal elements; 2) constructing a minimum distance constraint term of an end-member spectrum; 3) constructing an abundance mixed norm sparsity constraint term; 4) constructing a sparsity constraint term of the gradient domain group of the abundance graph; 5) establishing a non-negative matrix unmixing model of space-spectrum combined multi-constraint optimization; 6) alternate direction iterativesolution; 7) outputting the terminal element and the abundance diagram obtained by the demultiplex. As the distance between the spectrum of the end element of the hyperspectral image and the geometrical cent of mass is small, the abundance sparsity and the slice smoothness are fully utilized, the searching space for solving the end element and the abundance is limited by multiple constraints, thelocal minimum is avoided, and the optimal solution is obtained by iterative solution; compared with the traditional non-negative matrix unmixing model method, the invention improves the precision of unmixing, enhances the robustness of the method to noise, and can be widely applied to the hyperspectral unsupervised unmixing in the fields of land resources, mineral exploration and precision agriculture.

Description

technical field [0001] The invention relates to remote sensing hyperspectral data processing technology, in particular to a space-spectrum joint multi-constraint optimization non-negative matrix unmixing method. Background technique [0002] Due to its spectral correlation and rich spatial information, hyperspectral data are widely used in military monitoring, precision agriculture, and mineral exploration. Among them, hyperspectral data unmixing is the key technology of quantitative remote sensing analysis. The basic principle of hyperspectral data unmixing is to decompose a single pixel spectrum into a combination of several pure pixel (end member) spectra. The theoretical basis is that due to the limitation of the spatial resolution of the imaging spectrometer, there are a large number of mixed pixels in the obtained hyperspectral image, and each mixed pixel contains a variety of pure substances (ie, end members). [0003] Many unmixing algorithms for hyperspectral data...

AI Technical Summary

Problems solved by technology

However, the above method only utilizes the correlation of hyperspectral spectral information, does not effectively combine spatial-spectral information, the unmixing accuracy is low, and the performance of the algorithm decreases when there is noise in the data

Images