Optimized Multiclass Fault Detection in Solar Photovoltaic Modules Using TinySqueezeNet With Minimal Computational Overhead

Abstract As solar photovoltaic (PV) energy continues to emerge as a vital renewable resource, ensuring its efficiency and reliability through early fault detection is paramount. The proposed model introduces an enhanced TinySqueezeNet model, specifically optimized for identifying and classifying faults in solar photovoltaic panels using thermal and electroluminescence imaging. The TinySqueezeNet model demonstrates outstanding performance in solar fault classification, achieving an optimal balance between feature compression and expansion through its lightweight fire-module-based architecture. It efficiently handles diverse classification tasks, from binary to 12-class scenarios, with minimal computational overhead. The model's exceptional accuracy, precision, recall, and F1 scores underscore its effectiveness, while receiver operating characteristic curves and confusion matrices validate its adaptability. TinySqueezeNet achieves high accuracy with significantly fewer parameters, offering superior efficiency compared to larger models. For instance, it achieved 96% accuracy in 2-class classification with 4.06 million parameters. In 12-class tasks, it reached 89% accuracy. This efficiency and adaptability make TinySqueezeNet a robust and scalable solution for real-world solar PV fault detection and classification tasks.