Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Enhancing Martian Terrain Recognition with Deep Constrained Clustering

Tejas Panambur, Mario Parente

TL;DR

The paper tackles Martian terrain recognition under challenging visual variation by introducing Deep Constrained Clustering with Metric Learning (DCCML), which uses soft and hard pairwise constraints to guide unsupervised clustering toward geologically meaningful terrain groups. It combines textures-aware embeddings from DEP or ViT with a PCCClustering framework and leverages spatial, depth, stereo (LR), and RSM constraints to enforce geological consistency, coupled with metric learning via triplet loss. On Curiosity Mastcam data, DCCML achieves tighter, more homogeneous clusters (DBI substantially lower) and higher retrieval accuracy than prior deep clustering approaches, demonstrating improved alignment with expert taxonomies. The approach has practical implications for autonomous rover exploration and site prioritization, with plans to extend to Perseverance data and to develop labeled datasets for Generalized Class Discovery, enabling robust analysis of unlabeled Martian imagery in future missions.

Abstract

Martian terrain recognition is pivotal for advancing our understanding of topography, geomorphology, paleoclimate, and habitability. While deep clustering methods have shown promise in learning semantically homogeneous feature embeddings from Martian rover imagery, the natural variations in intensity, scale, and rotation pose significant challenges for accurate terrain classification. To address these limitations, we propose Deep Constrained Clustering with Metric Learning (DCCML), a novel algorithm that leverages multiple constraint types to guide the clustering process. DCCML incorporates soft must-link constraints derived from spatial and depth similarities between neighboring patches, alongside hard constraints from stereo camera pairs and temporally adjacent images. Experimental evaluation on the Curiosity rover dataset (with 150 clusters) demonstrates that DCCML increases homogeneous clusters by 16.7 percent while reducing the Davies-Bouldin Index from 3.86 to 1.82 and boosting retrieval accuracy from 86.71 percent to 89.86 percent. This improvement enables more precise classification of Martian geological features, advancing our capacity to analyze and understand the planet's landscape.

Enhancing Martian Terrain Recognition with Deep Constrained Clustering

TL;DR

The paper tackles Martian terrain recognition under challenging visual variation by introducing Deep Constrained Clustering with Metric Learning (DCCML), which uses soft and hard pairwise constraints to guide unsupervised clustering toward geologically meaningful terrain groups. It combines textures-aware embeddings from DEP or ViT with a PCCClustering framework and leverages spatial, depth, stereo (LR), and RSM constraints to enforce geological consistency, coupled with metric learning via triplet loss. On Curiosity Mastcam data, DCCML achieves tighter, more homogeneous clusters (DBI substantially lower) and higher retrieval accuracy than prior deep clustering approaches, demonstrating improved alignment with expert taxonomies. The approach has practical implications for autonomous rover exploration and site prioritization, with plans to extend to Perseverance data and to develop labeled datasets for Generalized Class Discovery, enabling robust analysis of unlabeled Martian imagery in future missions.

Abstract

Martian terrain recognition is pivotal for advancing our understanding of topography, geomorphology, paleoclimate, and habitability. While deep clustering methods have shown promise in learning semantically homogeneous feature embeddings from Martian rover imagery, the natural variations in intensity, scale, and rotation pose significant challenges for accurate terrain classification. To address these limitations, we propose Deep Constrained Clustering with Metric Learning (DCCML), a novel algorithm that leverages multiple constraint types to guide the clustering process. DCCML incorporates soft must-link constraints derived from spatial and depth similarities between neighboring patches, alongside hard constraints from stereo camera pairs and temporally adjacent images. Experimental evaluation on the Curiosity rover dataset (with 150 clusters) demonstrates that DCCML increases homogeneous clusters by 16.7 percent while reducing the Davies-Bouldin Index from 3.86 to 1.82 and boosting retrieval accuracy from 86.71 percent to 89.86 percent. This improvement enables more precise classification of Martian geological features, advancing our capacity to analyze and understand the planet's landscape.

Paper Structure

This paper contains 32 sections, 5 equations, 17 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related work
  3. Self-Supervised Learning
  4. Deep Constrained Clustering
  5. Martian Terrain recognition
  6. Deep self-supervised texture recognition with constrained clustering
  7. Network Architectures
  8. Deep constrained clustering using metric learning
  9. Constraint-Based Clustering
  10. Generating Constraints between Neighbouring Patches
  11. Spatial Constraints
  12. Depth Constraints
  13. Soft Must-Link Constraints
  14. Establishing Constraints between Left and Right Image Patches
  15. Establishing Constraints between RSM Pairs
  16. ...and 17 more sections

Figures (17)

  • Figure 1:
  • Figure 2:
  • Figure 3:
  • Figure 4:
  • Figure 6:
  • ...and 12 more figures