Effectiveness of Rooftop Photovoltaic Inspection Based on RGB Aerial Imagery and YOLOv11 Compared with Conventional Inspection

Abstract

Conventional rooftop photovoltaic (PV) inspection is still commonly performed through direct visual assessment by technicians. However, this approach is time-consuming, exposes workers to higher safety risks, and often produces less consistent documentation. This study develops an automatic inspection approach based on RGB aerial imagery and deep learning and compares its effectiveness with conventional inspection. A case study was conducted on a 60 kWp rooftop PV system at the Smart and Green Learning Center (SGLC), Faculty of Engineering, Universitas Gadjah Mada. A total of 150 RGB images were acquired using a drone, annotated, and used to train a YOLOv11 model through a Roboflow workflow. Technical evaluation was complemented by a 24-item Likert survey involving five technicians and by investment and operational cost analysis. The model achieved an mAP@50 of 98.3%, precision of 100%, and recall of 97.0%. Automatic inspection reduced inspection time from approximately 95 minutes to 18 minutes per session, corresponding to a time saving of about 81%. Technician perception also favored the automatic method, with an average score of 4.82 compared with 3.34 for the conventional approach, and the difference was statistically significant (p = 0.000036). Although the automatic method requires higher initial investment, it yields lower annual operational costs. These findings indicate that drone-based RGB imaging and YOLOv11 provide a feasible approach for modern rooftop PV maintenance, particularly for periodic inspection and multi-installation deployment.