|
Abstract
|
This study presents a hybrid image processing and machine learning framework for automatic dust detection and temperature estimation on photovoltaic (PV) panels. The proposed pipeline combines fuzzy clustering for rough panel segmentation, an SVM classifier enhanced with intensity–texture features and fuzzy logic for dust classification, and a semi-empirical plus machine learning-based model for estimating the thermal coefficient α. Environmental parameters such as humidity, ambient and panel temperatures, pressure, and solar irradiance are integrated to refine predictions. Validation using field images and sensor data demonstrates high accuracy in dust detection and strong correlation in α estimation. Spatial temperature maps generated from the hybrid model effectively capture localized heating effects due to dust, demonstrating consistency with expected thermal behavior on PV panels. The system shows robust performance across lighting conditions and seasonal variations, offering a practical tool for real-time PV panel monitoring and predictive maintenance.
|