Abstract
The global installation capacity of solar photovoltaic (PV) systems is exponentially increasing. However, the accumulation of soil and debris on solar panels significantly reduces their efficiency, necessitating frequent cleaning to maintain optimal energy output. This study presents a deep learning-based approach for the recognition and classification of soiled PV images, aimed at enhancing the capabilities of solar cleaning robots through the HALCON software framework. Using EANN and CNN architecture along with advanced image processing techniques, the proposed system achieves precise detection and classification of soiling patterns. The HALCON framework facilitates image acquisition, preprocessing, segmentation, and deployment of trained models for robotic control. The trained models demonstrate exceptional accuracy, with the EANN and CNN achieving classification precision of 99.87% and 99.91%, respectively. Experimental results highlight the system's potential to improve automation of cleaning strategies, reduce unnecessary cleaning cycles, and enhance the overall performance of solar panels. This research underscores the transformative role of intelligent visual analysis in optimizing maintenance practices for renewable energy applications.