NanoDistillNet: a lightweight diagnostic model for PV module faults in UAV-based inspection
Xinyan Zhang
As photovoltaic (PV) power plants expand, UAV-based infrared thermography has become a standard tool for efficient and safe fault diagnosis. However, most deep learning models are too large and computationally heavy for real-time deployment on edge devices. We propose NanoDistillNet, a lightweight model based on knowledge distillation designed for on-board processing. The framework utilizes a teacher network incorporating a Frequency-Adaptive Attention Module (FAAM) and a Kolmogorov-Arnold Netwo