Automated Image Processing and Machine-Learning Based System for Efficient Dust Detection and Cleaning of Solar Panels

Abstract

Solar photovoltaic (PV) technology is a promising renewable energy source; however, its efficiency considerably reduces dust accumulation, particularly in arid or polluted environments. Traditional water-based or manual cleaning methods are inefficient, costly, and unsustainable. This study proposes an automated dry-cleaning system that integrates machine learning and image processing for real-time dust detection and targeted cleaning. The system employs Canny edge detection and YOLOv5 to isolate PV panel regions and classify dust severity, labeling panels with ≥30% surface coverage as “dusty.” Arduino Mega and Jetson Nano jointly control a brush-based mechanism driven by dual DC gear motors, optimized for vertical (400rpm) and horizontal (200rpm) cleaning. Experimental evaluation on 15-W PV modules demonstrated that the system improved module efficiency from 77.1% under dusty conditions to 96.0% after cleaning, corresponding to an average 39% increase in output power. The dust detection algorithm achieved 82% classification accuracy, and the optimal cleaning performance was obtained with two sweeps (≈1min 59s), beyond which additional passes yielded negligible gains. By integrating accurate detection, optimized automation, and sustainable dry-cleaning, the proposed framework provides a scalable theresource-efficient solution for maintaining long-term PV performance.