"Spot the Glacial Lake'' with the magic of SAR and Deep Learning
收藏资源简介：
Description:This repository contains the dataset, model weights, and code for GLDeepNet, a fully automated deep convolutional neural network (DCNN) designed for accurate and scalable glacial lake mapping. The approach leverages publicly accessible satellite datasets—including multispectral, thermal, SAR (microwave), and digital elevation models (DEMs)—to enable efficient, cost-effective analysis of glacial lakes, particularly in high-mountain Asia. Key Features: Automated Deep Learning Model: GLDeepNet uses a 2D convolutional architecture with optimized hyperparameters for glacial lake segmentation. Comprehensive Pre-processing Pipeline: Includes optimal spectral band selection and SAR-specific transformations. Post-processing Enhancements: Morphological operations are applied to refine the detected lake boundaries. High Performance: Achieves an accuracy of 0.98, precision of 0.90, recall of 0.94, and R² of 0.91 compared to reference data. Robust Cross-Regional Testing: Trained on 1,050 images from the eastern Himalayas and tested on 750 images from the more vulnerable north-central Himalayan region. Applications:The dataset and model support large-scale glacial lake mapping and monitoring, offering critical insights for glacier retreat analysis and risk assessment of glacier lake outburst floods (GLOFs). Licensing and Citation:Please cite the accompanying publication or this Zenodo record when using the code or data in academic work. Contact:For questions or collaboration inquiries, contact Subhranil Mustafi at <subhranil_r@isical.ac.in>.
