HyDeMiC: A Deep Learning-based Mineral Classifier using Hyperspectral Data

TL;DR

HyDeMiC introduces a deep learning framework for robust mineral classification from hyperspectral data by training a 1D CNN on 444 USGS spectra representing 115 minerals and emulating AVIRIS sensor bands. The model encodes spectral signatures and maps to mineral labels via an encoder–decoder approach, enabling pixel-wise classification on 2D hyperspectral images. Across clean and noisy synthetic data, HyDeMiC achieves near-perfect performance ($MCC=1.00$ on clean data, $MCC=0.999$ at 5% noise, and $MCC=0.92$ at 10% noise) with high confidence scores, demonstrating strong robustness to spectral distortions. The framework supports practical mineral mapping and opens avenues for real-world deployment and broader hyperspectral classification applications, including uncertainty-aware extensions and applications beyond minerals.

Abstract

Hyperspectral imaging (HSI) has emerged as a powerful remote sensing tool for mineral exploration, capitalizing on unique spectral signatures of minerals. However, traditional classification methods such as discriminant analysis, logistic regression, and support vector machines often struggle with environmental noise in data, sensor limitations, and the computational complexity of analyzing high-dimensional HSI data. This study presents HyDeMiC (Hyperspectral Deep Learning-based Mineral Classifier), a convolutional neural network (CNN) model designed for robust mineral classification under noisy data. To train HyDeMiC, laboratory-measured hyperspectral data for 115 minerals spanning various mineral groups were used from the United States Geological Survey (USGS) library. The training dataset was generated by convolving reference mineral spectra with an HSI sensor response function. These datasets contained three copper-bearing minerals, Cuprite, Malachite, and Chalcopyrite, used as case studies for performance demonstration. The trained CNN model was evaluated on several synthetic 2D hyperspectral datasets with noise levels of 1%, 2%, 5%, and 10%. Our noisy data analysis aims to replicate realistic field conditions. The HyDeMiC's performance was assessed using the Matthews Correlation Coefficient (MCC), providing a comprehensive measure across different noise regimes. Results demonstrate that HyDeMiC achieved near-perfect classification accuracy (MCC = 1.00) on clean and low-noise datasets and maintained strong performance under moderate noise conditions. These findings emphasize HyDeMiC's robustness in the presence of moderate noise, highlighting its potential for real-world applications in hyperspectral imaging, where noise is often a significant challenge.

Figures (6)

• Figure 1: Figure: HyDeMiC framework for mineral classification using hyperspectral data and a 1D Convolutional Neural Network (CNN) architecture. The top section illustrates the training workflow, in which 1D spectral signatures from minerals are processed by a CNN comprising two convolutional layers (with kernel size 7 and increasing filter depth), max-pooling, dropout for regularization, and fully connected layers. The model is trained to predict encoded labels of 115 mineral types using an encoder-decoder strategy and a loss function computed against the encoded target labels. The bottom section depicts the prediction workflow, in which the trained 1D CNN model takes 1D spectral signatures or 2D hyperspectral images to generate encoded mineral predictions. These encoded outputs are decoded to produce human-readable mineral names, which are subsequently visualized as a 2D mineral distribution map.
• Figure 2: a) 1D reflectance spectral signature of Cuprite, Malachite, and Chalcopyrite, b) Generation of 1D synthetic spectral data with 1-10% of random noise added to the 1D spectral signature for the selected minerals, and c) schematic illustration of 2D synthetic hyperspectral images created using 1D synthetic spectral data by assigning them to each pixel for evaluating HyDeMiC's prediction capability.
• Figure 3: Training and validation loss for the developed HyDeMiC model.
• Figure 4: HyDeMiC prediction results on clean synthetic hyperspectral data. (a) Spatial distribution of mineral classes predicted by HyDeMiC, (b) Pixel-wise classification summary showing the total number of correct and incorrect predictions, (c) Prediction confidence as a function of pixel index, where each point represents an individual pixel colored by correctness, and (d) Density-normalized histogram of prediction confidence.
• Figure 5: HyDeMiC prediction results on 5% noisy synthetic hyperspectral data. (a) Spatial distribution of mineral classes predicted by HyDeMiC, (b) Pixel-wise classification summary showing the total number of correct and incorrect predictions, (c) Prediction confidence as a function of pixel index, where each point represents an individual pixel colored by correctness, and (d) Density-normalized histogram of prediction confidence.