Application and Evaluation of Deep Neural Networks for Airborne Hyperspectral Remote Sensing Mineral Mapping: A Case Study of the Baiyanghe Uranium Deposit in Northwestern Xinjiang, China

Abstract

Deep learning is a popular topic in machine learning and artificial intelligence research and has achieved remarkable results in various fields. In geological remote sensing, mineral mapping is an appealing application of hyperspectral remote sensing for geological surveyors. Whether deep learning can improve the mineral identification ability in hyperspectral remote sensing images, especially for the discrimination of spectrally similar and intimately mixed minerals, needs to be evaluated. In this study, shortwave airborne spectrographic imager (SASI) hyperspectral images of the Baiyanghe uranium deposit in Northwestern Xinjiang, China, were used as experimental data. Three deep neural network (DNN) models were designed: a fully connected neural network (FCNN), a one-dimensional convolutional neural network (1D CNN), and a one-dimensional and two-dimensional convolutional neural network (1D and 2D CNN). A sample dataset containing five minerals was constructed for model training and validation, which was divided into training, validation and test sets at a ratio of 6:2:2. The final test accuracies of the FCNN, 1D CNN, and 1D and 2D CNN were 91.24%, 93.67% and 94.77%, respectively. The three DNNs were used for mineral identification and mapping of SASI hyperspectral images of the Baiyanghe uranium mining area. The mapping results were compared with the mapping results of the support vector machine (SVM) and the mixture-tuned matched filtering (MTMF) method. Combined with the ground spectral data obtained by the spectrometer, spectral verification and interpretation were carried out on sections that the two kinds of methods identified differently. The verification results show that the mapping results of the 1D and 2D CNN were more accurate than those of the other methods. More importantly, for minerals with similar spectral characteristics, such as short-wavelength white mica and medium-wavelength white mica, the 1D and 2D CNN model had a more accurate discrimination effect than the other DNN models, indicating that the introduction of spatial information can improve the mineral identification ability in hyperspectral remote sensing images. In general, CNNs have good application prospects in geological mapping of hyperspectral remote sensing images and are worthy of further development in future work.

Research on Classification of Grassland Degeneration Indicator Objects Based on UAV Hyperspectral Remote Sensing and 3D_RNet-O Model

“…By combining 1D-CNN and 2D-CNN, scholars have simultaneously extracted spectral and spatial features, effectively enhancing feature extraction capabilities. However, the process is cumbersome and the improvement in classification accuracy is limited [ 35 ]. To address problems such as redundant information in adjacent spectral bands of hyperspectral images, incomplete information feature extraction between spectral dimensions in 2D-CNN, and high computational complexity of 3D-CNN, scholars have proposed a fusion of vegetation indices and 3D-2D-CNN classification method, which effectively enhances feature extraction capabilities.…”

An Evaluation of Convolutional Neural Networks for Lithological Mapping Based on Hyperspectral Images

“…In this aspect, 3D CNN [46], fully convolutional network [47], convolutional capsule network [48], and multiple deep learning or dual-channel frameworks such as ConvLSTM [49,50], 1D-2D CNN [51], 3D-2D CNN [52,53], convolutional autoencoder [54], graph convolutional network [55], CNN-local discriminant embedding [56], CNN-SAE [57], CNN-transformer learning [58], and joint attention network [59] have been introduced for spectralspatial hyperspectral classification and achieved state-ofthe-art performance. The combined use of CNNs and hyperspectral data provides the potential to assist in geological mapping involving mineral exploration [60][61][62][63]. On this account, this study focused on using CNNs to exploit hyperspectral data for lithological mapping.…”

Evaluating the Performance of Machine Learning and Deep Learning Techniques to HyMap Imagery for Lithological Mapping in a Semi-Arid Region: Case Study from Western Anti-Atlas, Morocco

“…Following their high capability in determining complex decision boundaries, several studies have also been conducted to access different SVM types for classifying remotely sensed data. Recently, deep learning and SVMs are the commonly used techniques in remote sensing data classification [21,[26][27][28]. Shirmard et al [24] used ML and DL for lithological mapping in a mineral-rich zone.…”