Swin-Transformer-Based CNN for Multi-Class PV Cell Anomaly Classification from Aerial Images
Year: 2026
Type: article
Abstract: Photovoltaic (Solar) panels, i.e., PV cells, are the most effective and economical renewable energy source generator. However, their power generation efficiency depends heavily on the cleanliness of their surfaces. The dust, snow, and bird drop cover the panel surface, reducing energy generation capability. Additionally, extreme weather causes electrical and physical damage to panels. This paper presents an approach for identifying these anomalies in aerial images using an advanced convolutional neural network (CNN). The proposed network uses the Swin Transformer (ST) to address the critical challenge of the traditional CNN’s interpretability. Multiple features were extracted using a shifted-patch-based multi-head self-attention at various stages, and these features were classified into six anomaly classes. The results show that the model succeeded with 92.53% accuracy and 92.64% F1-score. Thus, the proposed approach provides a practical solution for analysing renewable energy infrastructure from an aerial perspective.
Institution: Prince Mohammad bin Fahd University
Language: English
Subfield: Industrial and Manufacturing Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W7166323467