Drone-based solar panel inspection using machine learning

The paper represents the experimental implementation of a Drone-based solar panel inspection using YOLOv8-based object detection with Ultraviolet sensing for automated and enhanced defect inspection. A quadcopter platform was equipped with imaging sensors that was used to capture aerial images of the solar panels. The dataset used consists of 1500 annotated images categorized into clean panels, surface cracks, dust accumulation and thermal defects. The YOLOv8 model was fine-tuned using the dataset which had an input of 500x500 resolution for training a series of 100 epochs. Transfer learning enabled object localization and classification and RGB-based detection, for effective detection and identification of abnormal surfaces, discharge related anomalies where identified successfully that cannot be identified through standard imaging. Experimental validation also demonstrated reliable detection across all categories of defects and the results validated the completion of the project using lightweight deep learning model for object detection with multiple sensor UAV platform for cost-effective inspection, making a cost-efficient and automated solar farm monitoring possible and much more efficient than traditional methods.