Surface anomaly detection on island-based PV panels using edge neural networks

Surface anomaly detection on photovoltaic （PV） panels is crucial for their operation and maintenance， especially in island environments where challenges such as small anomaly sizes and minimal color differences are prevalent. Due to the poor accuracy and low efficiency of existing detection methods， the paper proposes a surface anomaly detection method for island-based PV panels using edge neural networks. First， by use of convolutional neural networks （CNNs） and attention mechanisms， an anomaly detection model， characterized by multi-scale feature fusion， is constructed to explore the features of fine-grained anomalies， thereby enhancing the surface anomaly detection accuracy. Additionally， a dual dynamic model compression technique is employed to reduce redundant channels and feature blocks， significantly lowering the model’s computational complexity and enabling rapid and accurate anomaly detection. The proposed method demonstrates strong performance in surface anomaly detection on PV panels in Zhoushan， highlighting its effectiveness and superiority.