Thermal and chromatic analysis for scalable photovoltaic hotspot detection

This study introduces a lightweight, interpretable framework for real-time PV fault detection using infrared (IR) thermography and perceptually uniform Lab* color space analysis. Unlike conventional methods reliant on high-dimensional texture or deep learning features, the proposed approach extracts just 80 statistical descriptors per image focused on luminance and chromaticity via patch-wise segmentation of IR thermographs. A suite of shallow classifiers (SVM, KNN, Decision Tree, Naive Bayes, and Ensemble) was trained on Lab*-derived features, achieving up to 95.2 % testing accuracy with sub-6-second training latency. SVM and KNN demonstrated superior diagnostic precision across healthy, hotspot, and faulty PV panels, validating the framework’s robustness and edge-computing compatibility. Beyond technical performance, the study quantifies the energy and climate impact of undetected hotspots in a 42.24 kW rooftop PV system. Results show a 17,620 kWh annual energy loss and a 27.6 % reduction in CO 2 mitigation potential equivalent to 20 fewer barrels of crude oil displaced. These findings underscore the urgency of scalable fault detection for sustainable solar deployment.