Zz G.

Key Skills

Software

No software listed

Top Subject Matter

AI-assisted InSAR phase denoising and mining subsidence monitoring

Mining subsidence monitoring using Sentinel-1 InSAR and UAV DEM integration

Top Data Types

Image

Top Task Types

Segmentation

Freelancer Overview

Core strengths include PyTorch, U-Net basics, and SNAP. Education includes Master of Science Candidate, China University of Mining and Technology-Beijing (2024). AI-training focus includes data types such as Geospatial and Tiled Imagery and labeling workflows including Segmentation, Evaluation, and Rating.

Labeling Experience

Mining Subsidence Monitoring Using Sentinel-1 InSAR and UAV DEM (2024-Present)

Processed Sentinel-1 TOPS SAR data and produced deformation products to support mining subsidence interpretation. Performed steps including coherence loss focus, phase filtering, unwrapping reliability checks, and geocoding accuracy as part of a monitoring research workflow. Integrated UAV-derived DEM/point-cloud products for local validation of deformation patterns and produced deformation maps and profile analysis.• Generated deformation maps and profile analyses for subsidence interpretation from geospatial SAR outputs.• Performed data-quality related evaluation of phase unwrapping reliability and filtering performance.• Supported cross-source validation by integrating UAV DEM/point-cloud with InSAR-derived deformation.• Produced technical summaries and research deliverables based on processed monitoring results.

2024 - Present

AI-Assisted InSAR Phase Denoising for High-Gradient Subsidence Areas (2024-Present)

Segmentation

Built a prototype AI workflow to improve InSAR phase quality for high-gradient subsidence areas using simulated interferograms and coherence-noise modeling. Explored U-Net / Attention U-Net style models to predict denoising weights and protect deformation-edge information. Evaluated filtering outputs using quantitative RMSE and qualitative phase-continuity checks on interferograms.• Trained/evaluated models on interferogram-derived image-like inputs with noise and coherence-related artifacts.• Focused on phase-quality improvement rather than manual human annotation or labeling.• Used adaptive filtering informed by model predictions to enhance unwrapping reliability.• Assessed results via residue-count statistics and visual interpretation of filtered interferograms.

2024 - Present

Education

Work History