Machine learning based hyperspectral image analysis: A survey

Abstract

Hyperspectral sensors enable the study of the chemical and physical properties of scene materials remotely for the purpose of identification, detection, chemical composition analysis, and physical parameter estimation 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 surveys 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/anomaly detection, unmixing, and physical/chemical 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.

Quantifying physiological trait variation with automated hyperspectral imaging in rice

“…One critical distinction is that the previous study used fresh weight as the metric for biomass. As water constitutes 90% of the mass in the growing tissue of a plant cell (Chavarria and dos Santos, 2012) and hyperspectral reflectance is sensitive to water (Gewali et al, 2019), it is likely that dry biomass is hard to predict from HSI data collected during the period plants are alive. Instead, dry weight may be better predicted using an alternative high-throughput method such as RBG imaging that can readily capture morphological traits (e.g., top surface area) that are highly correlated with biomass (Pandey et al, 2017).…”

“…Methods such as partial least squares regression (PLSR) are needed to quantify traits from HSI data while the recent use of machine learning (ML) algorithms to HSI-derived data has been shown to vastly improve predictions (Mir et al, 2019;Mishra et al, 2020;Arias et al, 2021). For example, support vector machines (SVM) have been recognized as one of the effective imaging classification algorithms (Noble, 2006;Gewali et al, 2019). Despite the capacity of PLSR and ML algorithms to handle large volumes of data, wavelength selection prior to building prediction models remains critical (Gewali et al, 2019;Yu et al, 2020).…”

“…For example, support vector machines (SVM) have been recognized as one of the effective imaging classification algorithms (Noble, 2006;Gewali et al, 2019). Despite the capacity of PLSR and ML algorithms to handle large volumes of data, wavelength selection prior to building prediction models remains critical (Gewali et al, 2019;Yu et al, 2020). Principal component analysis (PCA) is one example of an unsupervised method to select wavelengths.…”

Learning the Link between Albedo and Reflectance: Machine Learning-Based Prediction of Hyperspectral Bands from CTX Images

“…In the last two decades, machine learning has become an important tool in the analysis of remotely sensed data on Earth and the planetary sciences in general. Applications on Earth include land cover classification [5], target detection, unmixing and physical/chemical parameter estimation, employing a wide variety of approaches and model architectures [6]. On the Moon, machine-learning based systems for the automated detection of small craters is becoming an important tool to investigate the age and composition of lunar surfaces [7,8].…”

Classification of municipal solid waste using deep convolutional neural network model applied to multispectral images

“…Imaging methods are often used for recognizing material components in various classification tasks and should be evaluated for online waste monitoring considering that the operation of the WtE plant is highly dependent on waste fraction categories. [6][7][8] Of particular interest are deep convolution neural network (CNN) methods for image classification that have proved to outperform traditional machine-learning techniques such as logistic regression or support vector machines. [8][9][10][11] Image classification with deep CNN is a supervised machine learning method and is based on applying image convolutions with learnable filters to extract features relevant for image categories.…”