Hierarchical Spatial-Spectral Feature Extraction with Long Short Term Memory (LSTM) for Mineral Identification Using Hyperspectral Imagery.

Deep learning models are widely employed in hyperspectral image processing to integrate both spatial features and spectral features, but the correlations between them are rarely taken into consideration. However, in hyperspectral mineral identification, not only the spectral and spatial features of minerals need to be considered, but also the correlations between them are crucial to further promote identification accuracy. In this paper, we propose hierarchical spatial-spectral feature extraction with long short term memory (HSS-LSTM) to explore correlations between spatial features and spectral features and obtain hierarchical intrinsic features for mineral identification. In the proposed model, the fusion spatial-spectral feature is primarily extracted by stacking local spatial features obtained by a convolution neural network (CNN)-based model and spectral information together. To better exploit spatial features and spectral features, an LSTM-based model is proposed to capture correlations and obtain hierarchical features for accurate mineral identification. Specifically, the proposed model shares a uniform objective function, so that all the parameters in the network can be optimized in the meantime. Experimental results on the hyperspectral data collected by the Airborne Visible/Infrared Imaging Spectrometer (AVIRIS) in the Nevada mining area show that HSS-LSTM achieves an overall accuracy of 94.70% and outperforms other commonly used identification methods.

A Combined Quantitative Evaluation Model for the Capability of Hyperspectral Imagery for Mineral Mapping.

To analyze the influence factors of hyperspectral remote sensing data processing, and quantitatively evaluate the application capability of hyperspectral data, a combined evaluation model based on the physical process of imaging and statistical analysis was proposed. The normalized average distance between different classes of ground cover is selected as the evaluation index. The proposed model considers the influence factors of the full radiation transmission process and processing algorithms. First- and second-order statistical characteristics (mean and covariance) were applied to calculate the changes for the imaging process based on the radiation energy transfer. The statistical analysis was combined with the remote sensing process and the application performance, which consists of the imaging system parameters and imaging conditions, by building the imaging system and processing models. The season (solar zenith angle), sensor parameters (ground sampling distance, modulation transfer function, spectral resolution, spectral response function, and signal to noise ratio), and number of features were considered in order to analyze the influence factors of the application capability level. Simulated and real data collected by Hymap in the Dongtianshan area (Xinjiang Province, China), were used to estimate the proposed model's performance in the application of mineral mapping. The predicted application capability of the proposed model is consistent with the theoretical analysis.

Improved Classification Method Based on the Diverse Density and Sparse Representation Model for a Hyperspectral Image.

To solve the small sample size (SSS) problem in the classification of hyperspectral image, a novel classification method based on diverse density and sparse representation (NCM_DDSR) is proposed. In the proposed method, the dictionary atoms, which learned from the diverse density model, are used to solve the noise interference problems of spectral features, and an improved matching pursuit model is presented to obtain the sparse coefficients. Airborne hyperspectral data collected by the push-broom hyperspectral imager (PHI) and the airborne visible/infrared imaging spectrometer (AVIRIS) are applied to evaluate the performance of the proposed classification method. Results illuminate that the overall accuracies of the proposed model for classification of PHI and AVIRIS images are up to 91.59% and 92.83% respectively. In addition, the kappa coefficients are up to 0.897 and 0.91.