An Ensemble-Based Deep Learning Framework for Efficient Soiling Detection on Photovoltaic Panels

Abstract

Solar energy plays a pivotal role in renewable energy systems; however, dust accumulation on photovoltaic panels substantially reduces energy production effi ciency. Manual cleaning methods at large-scale plants are costly and impractical, highlighting the need for automated detection techniques. This study presentsa novel image processing and deep learning-based approach to accurately detect dusty PV panels. Images underwent preprocessing, including Hue, Saturation, Value color space conversion, and morphological operations to precisely segment dust-affected regions. Individual performances of DenseNet169, Xception, and InceptionV3 models were evaluated, and an ensemble model—Deep Solar Ensemble—was developed via soft voting. Experimental results demonstrated that the proposed ensemble achieved a superior classification accuracy of 97.02%, a precision of 97.29%, a recall of 96.56%, and an F1 score of 96.92% on a binary classification task. To address real-world applicability and robustness, the study was extended to include comparisons with lightweight architectures and testing on a more diverse, multi-class dataset containing various fault types, where the ensemble continued to show robust performance. The proposed methodology offers significant potential for automating solar panel maintenance, thereby enhancing operational efficiency, while also considering the trade offs between accuracy and computational cost for practical deployment.

Cite this article as: M. Balcı, A. Fındıkçı, M. Y. Erten, and H. Aydilek, “An ensemble-based deep learning framework for efficient soiling detection on photovoltaic panels,” Turk J Electr Power Energy Syst. Published online November 17, 2025. doi: 10.5152/tepes.2025.25020.