Detection of solar panels based on aerial images of the city of...
summary
Claims automated detection yields a citywide lower bound of 476 panels found across the sample area, about 2% of buildings, using high‑resolution aerial TIFFs (5 cm/px) and a building mask with a 0.5 m buffer. Model training annotates over 4,000 images and runs on AWS with Tiny YOLO (training ~10+ hours), achieving ~93% precision but frequent false positives from windows, roof patterns and shadows; some roofs split across tiles force dropping many buildings, so coverage is incomplete.
AI-written summary · about
chapters (15)
- 0:00Solar panel detection project
- 0:49Solar panel data problems in Poland
- 2:33Solar panel data sources in Poland
- 4:00Prior solar panel detection projects
- 4:45Data acquisition and preprocessing
- 6:10Building layer challenges
- 8:20Technologies for detection
- 9:27YOLO detection model
- 11:00Image annotation for training
- 11:22Training results and metrics
- 12:31Model predictions and errors
- 13:57Solar panel distribution in Poznań
- 14:45Project summary and limitations
- 15:27Q&A on model and data
- 17:16Closing remarks
related talks
27mUsing AI for cheaper and more efficient solar power
11mUsing AI to Accelerate the Smart Green Transition
27mUsing Few Shot Object Detection for Utility Pole Detection
28mGeoAI: Evolution of AI-Based Tree/Vegetation Detection Models Using Imagery
22mUrban Artificial Intelligence in Amsterdam
36mClassification of Satellite Imagery
24mNatural Disaster Damage Assessment from Satellite Imagery
01h 00mCloud and Shadow Recognition in Satellite Imagery