Dual-Spectral Spatial-Frequency Attention Network for Real-Time Photovoltaic Module Fault Detection

Photovoltaic (PV) modules are susceptible to internal faults (e.g., hot spots, open circuits) and external interference (e.g., shadows, vegetation coverage) during operation, leading to significant power efficiency degradation and reduced lifespan. This study proposes an enhanced YOLOv11 algorithm with a Spatial-Frequency Attention Module (SFAM) for real-time photovoltaic (PV) fault detection. By integrating Fast Fourier Transform (FFT) and adaptive convolution, SFAM optimizes multi-scale feature fusion, achieving 91.79% mAP@50 for infrared and 68.48% mAP@50 for visible-light detection on a dual-spectral dataset. The model attains 45 FPS with 9.11 GB memory usage, significantly improving small-target precision (e.g., <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$+4 \%$</tex> for bird excrement) while balancing accuracy and efficiency for drone-based PV inspections.