Research on Thermal Spot Fault Detection of Photovoltaic Modules Based on Improved YOLOv5s

Aiming at the problems of low precision and low efficiency of traditional PV module hot spot detection methods, this study proposes a technical route to implement intelligent inspection of PV power stations using UAVs. This route has the ability to automatically collect and analyze PV module image data, and adopts a deep learning detection method based on the STA-TCN model for PV module hot spot fault diagnosis, and realizes defect localization by solving the defect coordinates based on the camera POS data and the camera model. The STA-TCN model is a multivariate time-domain convolution network(TCN) feed-forward network model that introduces a multivariate TCN that integrates spatial and temporal attention mechanisms to address the failure of the traditional TCN to solve the problems of the traditional TCN. Feedforward network model to solve the problem that traditional TCN fails to fully consider the contribution of exogenous sequences to state prediction. Finally, the effectiveness of the model proposed in this study is demonstrated through comparative experiments. Based on the experimental results, the STA-TCN model shows higher performance and reliability in key metrics, such as detection accuracy, recall and mAP, compared with traditional machine learning methods.