On July 21, 2026, Wajiha Yasmeen will present her internship results on ” Detection of Off-The-Road Tyre Dumpsites at Mining Operations Using Sentinel-2 Imagery and Deep Learning Segmentation ” at 12:00 at the seminar room 3 in John-Skilton-Str. 4a.
From the abstract: Discarded off-the-road (OTR) tyres at mining sites represent a persistent environmental hazard, yet their locations are poorly documented across large mining regions. This work develops a supervised semantic segmentation pipeline to detect OTR tyre dumpsites from Sentinel-2 multispectral imagery across mining operations in Chile. Per-mine Sentinel-2 composites were exported and candidate dump areas identified through spectral thresholding, then manually verified and digitised in QGIS. A DeepLabV3+ network with a pretrained encoder was trained on 10-band input using a combined Dice and weighted binary cross-entropy loss. Results indicate that site-level detection is achievable with the current approach, while accurate delineation of dump extent remains constrained by the resolution of the dataset.
From the abstract: Discarded off-the-road (OTR) tyres at mining sites represent a persistent environmental hazard, yet their locations are poorly documented across large mining regions. This work develops a supervised semantic segmentation pipeline to detect OTR tyre dumpsites from Sentinel-2 multispectral imagery across mining operations in Chile. Per-mine Sentinel-2 composites were exported and candidate dump areas identified through spectral thresholding, then manually verified and digitised in QGIS. A DeepLabV3+ network with a pretrained encoder was trained on 10-band input using a combined Dice and weighted binary cross-entropy loss. Results indicate that site-level detection is achievable with the current approach, while accurate delineation of dump extent remains constrained by the resolution of the dataset.
