HotSD-Net: Convolutional Neural Network for Estimation of Hot Spots in Photovoltaic Modules via Infrared Imaging

This work presents a novel approach for hot spot probability estimation in photovoltaic (PV) modules using infrared (IR) imaging and a deep neural network-based architecture called HotSD-Net. Hot spots, which are localized areas of excessive heating, can significantly reduce the efficiency and useful life of PV modules. Standard traditional methods for detecting hot spots are time-consuming and not cost-effective since they rely on manual inspection. To address this, we propose HotSD-Net, a deep learning architecture designed to automate and enhance hot spot probability estimation. Leveraging IR imaging data, our network incorporates image information to detect potential hot spots with high precision. The model is trained on a dataset of IR images, using supervised learning to map input images to hot spot probability distributions. Our results demonstrate that HotSD-Net performs well, offering robustness in detecting and estimating the likelihood of hot spot occurrences in PV modules with a confidence level of 0.95. This automated approach offers a scalable solution for real-time monitoring and maintenance of solar energy systems.