Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

A UNet Model for Accelerated Preprocessing of CRISM Hyperspectral Data for Mineral Identification on Mars

Priyanka Kumari, Sampriti Soor, Amba Shetty, Archana M. Nair

TL;DR

The paper tackles the bottleneck of preprocessing CRISM MTRDR hyperspectral data for Martian mineral identification by introducing a 1D UNet autoencoder that encapsulates smoothing, baseline correction, and continuum-like adjustments. Trained on augmented MICA spectra and paired with MICAnet for mineral classification, it achieves a substantial speedup—reducing $1.5$ hours to $5$ minutes for an $800\times800$ scene—while maintaining competitive accuracy. Ablation studies show deeper IV-B configurations offer the best trade-off between computational cost and accuracy, enhancing both classwise and groupwise mineral detection by roughly $5\%$ over traditional pipelines. The approach promises real-time mineral mapping on Mars and can extend to other planetary hyperspectral datasets, broadening the impact of automated spectral preprocessing in planetary science.

Abstract

Accurate mineral identification on the Martian surface is critical for understanding the planet's geological history. This paper presents a UNet-based autoencoder model for efficient spectral preprocessing of CRISM MTRDR hyperspectral data, addressing the limitations of traditional methods that are computationally intensive and time-consuming. The proposed model automates key preprocessing steps, such as smoothing and continuum removal, while preserving essential mineral absorption features. Trained on augmented spectra from the MICA spectral library, the model introduces realistic variability to simulate MTRDR data conditions. By integrating this framework, preprocessing time for an 800x800 MTRDR scene is reduced from 1.5 hours to just 5 minutes on an NVIDIA T1600 GPU. The preprocessed spectra are subsequently classified using MICAnet, a deep learning model for Martian mineral identification. Evaluation on labeled CRISM TRDR data demonstrates that the proposed approach achieves competitive accuracy while significantly enhancing preprocessing efficiency. This work highlights the potential of the UNet-based preprocessing framework to improve the speed and reliability of mineral mapping on Mars.

A UNet Model for Accelerated Preprocessing of CRISM Hyperspectral Data for Mineral Identification on Mars

TL;DR

The paper tackles the bottleneck of preprocessing CRISM MTRDR hyperspectral data for Martian mineral identification by introducing a 1D UNet autoencoder that encapsulates smoothing, baseline correction, and continuum-like adjustments. Trained on augmented MICA spectra and paired with MICAnet for mineral classification, it achieves a substantial speedup—reducing hours to minutes for an scene—while maintaining competitive accuracy. Ablation studies show deeper IV-B configurations offer the best trade-off between computational cost and accuracy, enhancing both classwise and groupwise mineral detection by roughly over traditional pipelines. The approach promises real-time mineral mapping on Mars and can extend to other planetary hyperspectral datasets, broadening the impact of automated spectral preprocessing in planetary science.

Abstract

Accurate mineral identification on the Martian surface is critical for understanding the planet's geological history. This paper presents a UNet-based autoencoder model for efficient spectral preprocessing of CRISM MTRDR hyperspectral data, addressing the limitations of traditional methods that are computationally intensive and time-consuming. The proposed model automates key preprocessing steps, such as smoothing and continuum removal, while preserving essential mineral absorption features. Trained on augmented spectra from the MICA spectral library, the model introduces realistic variability to simulate MTRDR data conditions. By integrating this framework, preprocessing time for an 800x800 MTRDR scene is reduced from 1.5 hours to just 5 minutes on an NVIDIA T1600 GPU. The preprocessed spectra are subsequently classified using MICAnet, a deep learning model for Martian mineral identification. Evaluation on labeled CRISM TRDR data demonstrates that the proposed approach achieves competitive accuracy while significantly enhancing preprocessing efficiency. This work highlights the potential of the UNet-based preprocessing framework to improve the speed and reliability of mineral mapping on Mars.
Paper Structure (9 sections, 15 equations, 6 figures, 2 tables)

This paper contains 9 sections, 15 equations, 6 figures, 2 tables.

Table of Contents

  1. Introduction
  2. Methedology
  3. Data Preparation
  4. Model Architecture
  5. Model and Training Specifications
  6. Ablation Study
  7. Performance Analysis and Comparison with Traditional Preprocessing Pipeline
  8. Mineral Mapping on the Martian Surface
  9. Discussion and Conclusion

Figures (6)

  • Figure 1: The U-net architecture used in this study. The architecture (IV-B) is fixed by an ablation study which is detailed in section \ref{['sec:ablationstudy']}.
  • Figure 2: (a) Standard deviation of fluctuation noises over the epochs; (b) frequency of different mineral classes having top K proportions in the mixed spectra; (c) Distribution of top 3 proportions of the combined spectra in the dataset. These distributions were observed during the training of the best architecture (IV-B) and were consistent across all architectures.
  • Figure 3: Training and validation loss over epochs during the training of the best architecture IV-B.
  • Figure 4: Some sample synthetic spectra, and their preprocessing results by architecture IV-B compared to the ground truth.
  • Figure 5: Classwise and group-wise accuracy of the minerals considered in the experiment shows detection improvement of around 5% if the proposed preprocessing model is used than the framework.
  • ...and 1 more figures