Image Processing and Machine Learning for Hyperspectral Unmixing - An Overview and the HySUPP Python Package

Aug 2, 2026

Spectral pixels in hyperspectral imaging often represent mixtures of pure material spectra, or endmembers, due to limited spatial resolution, multiple scattering, and intimate mixing. Unmixing estimates endmember fractional abundances and can follow supervised, semisupervised, or unsupervised linear approaches depending on available prior knowledge. The work surveys both conventional and advanced unmixing methods, compares them across categories, and evaluates performance on three simulated and one real dataset. Results highlight scenario-dependent advantages, and an open-source HySUPP Python package enables reproduction.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages29

Reading Time1 h 13 min

AccessFree to download

Introduction

Hyperspectral unmixing motivation and mixing models

Linear and bilinear mixing assumptions

Nonlinearity sources and variability handling

Endmember definition and subjective unmixing implications

Image Processing and Machine Learning for Hyperspectral Unmixing: An Overview and the HySUPP Python Package Behnood Rasti, Senior Member, IEEE, Alexandre Zouaoui, Student Member, IEEE, Julien Mairal, Senior Member, IEEE and Jocelyn Chanussot, Fellow, IEEE arXiv :2308 .09375v2 [ ee ss .IV] 6 Oct 2023 Abstract—Spectral pixels are often a mixture of the pure spectra of the materials, called endmembers, due to the low spatial resolution of hyperspectral sensors, double scattering, and intimate mixtures of materials in the scenes. Unmixing estimates the fractional abundances of the endmembers within the pixel. Depending on the prior knowledge of endmembers, linear unmixing can be divided into three main groups: supervised, semisupervised, and unsupervised (blind) linear unmixing. Advancesin image processing and machine learning substantially affected unmixing. This paper provides an overview of advanced and conventional unmixing approaches. Additionally, we draw a critical comparison between advanced and conventional techniques from the three categories. We compare the performance of the unmixing techniques on three simulated and one real dataset. The experimental results reveal the advantages of different unmixing categories for different unmixing scenarios. Moreover, we provide an open-source Python-based package available at [https://github.com/BehnoodRasti/HySUPP](https://github.com/BehnoodRasti/HySUPP) to reproduce the results. Index Terms—Hyperspectral, unmixing, endmember extraction, abundance estimation, linear mixture, machine learning, deep learning, optimization. I. INTRODUCTION SPECTRAL unmixing is a crucial processing technique in hyperspectral remote sensing that can play a vital role in various fields such as mineral exploration, agriculture and crop monitoring, environmental monitoring, urban planning, remote sensing of planetary surfaces, pollution monitoring, medical imaging, water quality assessment, etc. The ability to separate and identify different materials in an image is made possible by the contiguous spectra captured by hyperspectral sensors. Through the use of endmembers, which are the unique spectral signatures of materials, unmixing algorithms can decompose the mixed spectral data into its constituent parts. However, due to low spatial resolution, multiple scattering, and intimate mixing, the measured spectrum within a pixel is generally a complex mixture of the pure spectra of the constituent materials, making unmixing a challenging task. Fig. 1 demonstrates how the reflectance of a mixed pixel captured by an optical hyperspectral camera is composed of Behnood Rasti (corresponding author) is with Helmholtz-Zentrum DresdenRossendorf, Helmholtz Institute Freiberg for Resource Technology, Machine Learning Group, Chemnitzer Straße 40, 09599 Freiberg, Germany; [b.rasti@hzdr.de](b.rasti@hzdr.de), [behnood.rasti@gmail.com](behnood.rasti@gmail.com) Alexandre Zouaoui, Jocelyn Chanussot, and Julien Mairal are with Univ. Grenoble Alpes, Inria, CNRS, Grenoble INP, LJK, 38000 Grenoble, France Manuscript received April 19, 2023; revised August 16, 2023 . two endmembers within that pixel. In hyperspectral remote sensing, a mixing model represents the observed spectral pixel as a function of the endmembers and their corresponding fractional abundances within the pixel’s area. Unmixing is the process of estimating the fractional abundances, either by estimating or extracting the endmembers or by relying on a library of endmembers. It may also involve determining the number of endmembers present. The mixing model is either linear or nonlinear, depending on the interaction of the incident light and the materials in the scene or sample. In linear unmixing, the endmembers are assumed to belinearly mixed, which is valid when each light ray interacts with only one material before reaching the sensor, as shown in Figure 2 (a) . This assumption is common in Earth observation applications where mac

FAQs

Why is hyperspectral unmixing challenging in practice?

Because each pixel may contain complex mixtures caused by low spatial resolution, multiple scattering, and intimate mixing, so the observed spectrum is not simply the pure endmembers.

How do supervised, semisupervised, and unsupervised linear unmixing differ?

They differ based on what prior knowledge of endmembers is available when estimating fractional abundances, ranging from using labeled endmembers to blind estimation.

What distinguishes linear mixing from bilinear mixing in the modeling assumptions?

Linear mixing assumes each light ray interacts with only one material before reaching the sensor, while bilinear mixing assumes double scattering or interaction with two materials.

Aug 2, 2026

Upload by Lucas Martin