Direct Head model estimates rooftop solar capacity from aerial imagery without area conversion
A French study reports a segmentation-guided architecture that regresses photovoltaic capacity directly, cutting error against an area-density baseline on 7,590 installations.
In plain words
Planners need to know how much rooftop solar power exists, but official records can be incomplete. The researchers built a model that looks at aerial images and predicts each roof's capacity directly, instead of first measuring panel area and multiplying by a power-per-area figure. On French rooftop data, this direct approach was more accurate than the conventional two-step method.
What was announced
Distribution planners want installation-level inventories of rooftop photovoltaic capacity, but interconnection records may be incomplete or aggregated by location, according to the paper's abstract. Conventional image-based estimators segment solar arrays, convert the resulting mask into physical area, and multiply by an estimated power density, which ties capacity prediction to mask-derived area. The authors propose Direct Head, an architecture that keeps U-Net segmentation and soft-mask pooling but regresses nonnegative capacity directly from PV-region features and area descriptors.
The team evaluated three independent runs on 7,590 French rooftop installations, using both the original split and a department-based holdout. Relative to a parameter-matched area-density neural model, Direct Head reduced mean absolute error by 8.7% and 12.5%, and symmetric mean absolute percentage error by 14.0% and 20.7%, respectively. The authors report that comparisons with conventional estimators, controlled architecture variants, and auxiliary-loss ablations show segmentation provides useful spatial evidence without capacity needing to come from an area-density product.
What it does not show
- Evidence is limited to the paper's abstract; full experimental details, baselines, and peer-review status are not available in the supplied source.
- Results cover 7,590 French rooftop installations, and the abstract does not establish performance in other countries or on other imagery datasets.
- The reported error reductions come from the authors' own evaluation and have not been independently confirmed.
For security and IT teams
- Scope
- The work targets installation-level rooftop PV capacity inventories for distribution planning where interconnection records are incomplete or spatially aggregated.
Sources
Source-support review completed 19 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.
Coverage: arXiv: abstract.
Updates and corrections
No corrections so far.
First published 19 Nov 2026
New to the subject? AI from the Ground Up starts from the beginning.