[Paper Review] Machine learning based hyperspectral image analysis: A survey

Utsav B. Gewali, Sildomar T. Monteiro+1 more|arXiv (Cornell University)|Feb 23, 2018

Remote-Sensing Image ClassificationEngineering207 references105 citations

TL;DR

A comprehensive survey of machine learning techniques applied to hyperspectral image analysis, organized by analysis task and ML algorithm, highlighting trends, challenges, and future directions.

ABSTRACT

Hyperspectral sensors enable the study of the chemical properties of scene materials remotely for the purpose of identification, detection, and chemical composition analysis of objects in the environment. Hence, hyperspectral images captured from earth observing satellites and aircraft have been increasingly important in agriculture, environmental monitoring, urban planning, mining, and defense. Machine learning algorithms due to their outstanding predictive power have become a key tool for modern hyperspectral image analysis. Therefore, a solid understanding of machine learning techniques have become essential for remote sensing researchers and practitioners. This paper reviews and compares recent machine learning-based hyperspectral image analysis methods published in literature. We organize the methods by the image analysis task and by the type of machine learning algorithm, and present a two-way mapping between the image analysis tasks and the types of machine learning algorithms that can be applied to them. The paper is comprehensive in coverage of both hyperspectral image analysis tasks and machine learning algorithms. The image analysis tasks considered are land cover classification, target detection, unmixing, and physical parameter estimation. The machine learning algorithms covered are Gaussian models, linear regression, logistic regression, support vector machines, Gaussian mixture model, latent linear models, sparse linear models, Gaussian mixture models, ensemble learning, directed graphical models, undirected graphical models, clustering, Gaussian processes, Dirichlet processes, and deep learning. We also discuss the open challenges in the field of hyperspectral image analysis and explore possible future directions.

Motivation & Objective

• Provide a broad overview of hyperspectral image analysis tasks and the ML methods used to address them.
• Compare and categorize ML algorithms with respect to land cover classification, target/anomaly detection, unmixing, and physical/chemical parameter estimation.
• Identify open challenges and propose directions for future research in ML-based hyperspectral analysis.
• Demonstrate the flexibility of ML methods over physics-based or fixed-model approaches in hyperspectral contexts.

Proposed method

• Organize methods by image analysis task and by ML algorithm type, and map tasks to suitable algorithm families.
• Survey 205 methods published in peer-reviewed venues and summarize them in a two-way task-algorithm mapping.
• Discuss dimensionality challenges, data variability, and the role of feature extraction and spatial-spectral information.
• Highlight open challenges and future directions in hyperspectralML research.

Experimental results

Research questions

• RQ1 What machine learning algorithms have been applied to each hyperspectral image analysis task (land cover classification, target/anomaly detection, unmixing, parameter estimation)?
• RQ2 How do different ML categories (Gaussian models, regression, SVM, GMMs, graphical models, clustering, Gaussian processes, Dirichlet processes, deep learning) perform across hyperspectral tasks?
• RQ3 What are the current challenges in hyperspectral ML (e.g., high dimensionality, limited ground truth, spectral variability) and potential future directions?

Key findings

• The survey covers 205 ML-based hyperspectral analysis methods across four tasks.
• There is a broad adoption of diverse ML approaches, including deep learning, Gaussian models, regression, SVMs, probabilistic models, and graphical models, applied to different tasks.
• Dimensionality reduction, band selection, and incorporation of spatial information are common strategies to address high dimensionality.
• Bayesian methods are highlighted for handling uncertainties in high-dimensional hyperspectral data.
• The paper identifies open challenges and suggests future research directions in ML for hyperspectral analysis.