Computer Vision Data Labeling for Landslide Motion Detection
Developed and managed a computer vision data labeling pipeline for landslide risk analysis using thousands of video frames converted to structured displacement data. Labeled vegetation in images for Random Forest training to enable vegetation segmentation accuracy under changing environmental conditions. Implemented preprocessing with Gaussian filtering and CLAHE to optimize image data quality for downstream model training. • Designed annotation workflows for pixel-wise segmentation of vegetation in slope images. • Managed the manual and automated labeling process across a 14,000-datapoint custom dataset. • Validated and refined segmentation labels to improve model accuracy prior to publication. • Coordinated integration of labeled data with machine learning pipelines for geoscience research.