Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

MCD-Net: A Lightweight Deep Learning Baseline for Optical-Only Moraine Segmentation

Zhehuan Cao, Fiseha Berhanu Tesema, Ping Fu, Jianfeng Ren, Ahmed Nasr

TL;DR

Problem: automated moraine segmentation from optical imagery is hampered by weak contrast and limited high-resolution DEMs. Approach: the authors present MCD-Net, a lightweight DeepLabV3+-style model with a MobileNetV2 backbone and CBAM, evaluated on a new optical-only Moraine Dataset (MCD) of 3,340 high-resolution images from Sichuan and Yunnan. Contributions: (i) public release of the 3,340-image dataset, (ii) a reproducible baseline achieving 62.3% mIoU and 72.8% Dice with a small parameter count and low computational cost, and (iii) ablation and cross-region analyses; (iv) a robust evaluation protocol. Findings: optical-only moraine segmentation is feasible and competitive, though ridge delineation remains constrained by sub-pixel widths and spectral ambiguity; cross-region generalization reveals non-trivial domain shifts. Significance: provides a scalable, deployable baseline and benchmark to advance high-altitude glacial monitoring with optical data.

Abstract

Glacial segmentation is essential for reconstructing past glacier dynamics and evaluating climate-driven landscape change. However, weak optical contrast and the limited availability of high-resolution DEMs hinder automated mapping. This study introduces the first large-scale optical-only moraine segmentation dataset, comprising 3,340 manually annotated high-resolution images from Google Earth covering glaciated regions of Sichuan and Yunnan, China. We develop MCD-Net, a lightweight baseline that integrates a MobileNetV2 encoder, a Convolutional Block Attention Module (CBAM), and a DeepLabV3+ decoder. Benchmarking against deeper backbones (ResNet152, Xception) shows that MCD-Net achieves 62.3% mean Intersection over Union (mIoU) and 72.8% Dice coefficient while reducing computational cost by more than 60%. Although ridge delineation remains constrained by sub-pixel width and spectral ambiguity, the results demonstrate that optical imagery alone can provide reliable moraine-body segmentation. The dataset and code are publicly available at https://github.com/Lyra-alpha/MCD-Net, establishing a reproducible benchmark for moraine-specific segmentation and offering a deployable baseline for high-altitude glacial monitoring.

MCD-Net: A Lightweight Deep Learning Baseline for Optical-Only Moraine Segmentation

TL;DR

Problem: automated moraine segmentation from optical imagery is hampered by weak contrast and limited high-resolution DEMs. Approach: the authors present MCD-Net, a lightweight DeepLabV3+-style model with a MobileNetV2 backbone and CBAM, evaluated on a new optical-only Moraine Dataset (MCD) of 3,340 high-resolution images from Sichuan and Yunnan. Contributions: (i) public release of the 3,340-image dataset, (ii) a reproducible baseline achieving 62.3% mIoU and 72.8% Dice with a small parameter count and low computational cost, and (iii) ablation and cross-region analyses; (iv) a robust evaluation protocol. Findings: optical-only moraine segmentation is feasible and competitive, though ridge delineation remains constrained by sub-pixel widths and spectral ambiguity; cross-region generalization reveals non-trivial domain shifts. Significance: provides a scalable, deployable baseline and benchmark to advance high-altitude glacial monitoring with optical data.

Abstract

Glacial segmentation is essential for reconstructing past glacier dynamics and evaluating climate-driven landscape change. However, weak optical contrast and the limited availability of high-resolution DEMs hinder automated mapping. This study introduces the first large-scale optical-only moraine segmentation dataset, comprising 3,340 manually annotated high-resolution images from Google Earth covering glaciated regions of Sichuan and Yunnan, China. We develop MCD-Net, a lightweight baseline that integrates a MobileNetV2 encoder, a Convolutional Block Attention Module (CBAM), and a DeepLabV3+ decoder. Benchmarking against deeper backbones (ResNet152, Xception) shows that MCD-Net achieves 62.3% mean Intersection over Union (mIoU) and 72.8% Dice coefficient while reducing computational cost by more than 60%. Although ridge delineation remains constrained by sub-pixel width and spectral ambiguity, the results demonstrate that optical imagery alone can provide reliable moraine-body segmentation. The dataset and code are publicly available at https://github.com/Lyra-alpha/MCD-Net, establishing a reproducible benchmark for moraine-specific segmentation and offering a deployable baseline for high-altitude glacial monitoring.
Paper Structure (34 sections, 8 equations, 10 figures, 5 tables)

This paper contains 34 sections, 8 equations, 10 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Related Work
  3. Classical Geomorphological Mapping
  4. Deep Learning in Glacial Geomorphology
  5. Research Gap in Moraine-Specific Identification
  6. Positioning of This Study
  7. Dataset and Evaluation Metrics
  8. Construction and Coverage
  9. Annotation Protocol
  10. Class Imbalance and Scale Diversity
  11. Partitioning and Preprocessing
  12. Evaluation Protocol and Metrics
  13. Methodology: The MCD-Net Architecture
  14. Motivation for a Lightweight Baseline
  15. Architecture Overview
  16. ...and 19 more sections

Figures (10)

  • Figure 1: Geographical distribution of moraine areas and representative examples from Sichuan and Yunnan.
  • Figure 2: Representative dataset examples under varying conditions: (a) strong shadows, (b) low-contrast rocky surfaces, (c) glacier-covered moraine deposits, (d) vegetation-covered moraines.
  • Figure 3: Pixel distribution across classes.
  • Figure 4: Histogram of moraine sizes (in pixels).
  • Figure 5: Proposed MCD-Net architecture. The pipeline consists of a MobileNetV2 backbone, a CBAM block for feature refinement, ASPP for multi-scale context, and a DeepLabV3+ decoder.
  • ...and 5 more figures