Dust Detection on Solar Photovoltaic Panels Used in Optoelectronics with Convolutional Neural Network-Based Deep Learning Models DOI Open Access
Fatih Uysal

International Journal of Computational and Experimental Science and Engineering, Journal Year: 2025, Volume and Issue: 11(1)

Published: Feb. 11, 2025

Solar photovoltaic panels, one of the optoelectronic device types, contain a large number cells. The maintenance these solar panels with cells is very important for efficiency energy obtained from panel. As time passes, dust may form on due to various weather conditions and environments where are located. In order maintain in timely manner increase efficiency, this study aims detect panels. For reason, an open source dataset consisting normal, clean well-maintained containing was used. Since amount small amounts classes unbalanced, firstly, data augmentation operations were performed make it balanced. use balanced classification phase deep learning models, divided into 80% training 20% testing. After process, total four models based convolutional neural networks, including MobileNetv1 detection ResNet three different layers, During processes, two optimization methods used train each model. result studies, highest accuracy value found be 0.993 model, which trained using AdamW method had 18 layers.

Language: Английский

Dust Detection on Solar Photovoltaic Panels Used in Optoelectronics with Convolutional Neural Network-Based Deep Learning Models DOI Open Access
Fatih Uysal

International Journal of Computational and Experimental Science and Engineering, Journal Year: 2025, Volume and Issue: 11(1)

Published: Feb. 11, 2025

Solar photovoltaic panels, one of the optoelectronic device types, contain a large number cells. The maintenance these solar panels with cells is very important for efficiency energy obtained from panel. As time passes, dust may form on due to various weather conditions and environments where are located. In order maintain in timely manner increase efficiency, this study aims detect panels. For reason, an open source dataset consisting normal, clean well-maintained containing was used. Since amount small amounts classes unbalanced, firstly, data augmentation operations were performed make it balanced. use balanced classification phase deep learning models, divided into 80% training 20% testing. After process, total four models based convolutional neural networks, including MobileNetv1 detection ResNet three different layers, During processes, two optimization methods used train each model. result studies, highest accuracy value found be 0.993 model, which trained using AdamW method had 18 layers.

Language: Английский

Citations

0