Date : 23/06/2025
Hour: 20:00
Student Name & ID: İrem Ezgi Bayar 245105112
Supervisor: Prof. Dr. Ömer KARAL
Topic: Power Plant Fuel Type Recognition via Image Quality-Aware Deep Learning
Link or Room: Meeting Link(Online) : https://meet.google.com/hhb-bmhu-ggb
Abstract: ‘This study tackles the challenge of fuel-type classification for global power plants using satellite imagery. Given the noisy, artifact-laden nature of raw satellite datasets—rife with clouds, uninformative regions, or absent infrastructure—we propose a multi-step data curation pipeline to refine input quality before model training. Images were acquired from the Global Power Plant Database using the Mapbox API and subjected to a three-tiered filtering system: (1) YOLOv10-based detection to retain power plant structures, (2) boringness analysis using color entropy and texture variance to eliminate visually monotonous scenes, and (3) validation-driven exclusion of images prone to overfitting or irrelevance.’
