RTPV-YOLO: Real-Time Photovoltaic Detection With UAV-Based Thermal and RGB Imaging

Unmanned Aerial Vehicles (UAVs) have become essential tools in photovoltaic (PV) defect detection. However, traditional convolutional neural network (CNN) models often experience significant performance degradation when detecting tiny hot spots in thermal images and irregularly shaped obstructions in RGB images, leading to missed and false detections. To address these challenges, we propose the Real-Time Photovoltaic YOLO network (RTPV-YOLO), specifically designed for real-time PV defect detection using UAVs. Built upon the YOLOv9 architecture, RTPV-YOLO integrates Spatial to Depth Convolution (SPD-Conv) and Convolutional Block Attention Module (CBAM) to enhance the detection of tiny thermally-induced hot spots by preserving fine-grained information and focusing on critical features. Additionally, we design a Reparameterization Adaptive Kernel Convolution Module (RepAKCM) for feature extraction and fusion within the RTPV-YOLO network. This module not only reduces the computational load and enhances the inference efficiency of the network, but also effectively handles objects of various shapes and sizes, such as bird droppings and leaves on PV panels. Experimental results on a custom-built PV defect dataset demonstrate that RTPV-YOLO achieves a mean Average Precision (mAP) of 91%, outperforming other comparable models. Furthermore, RTPV-YOLO can be effectively deployed on resource-constrained UAVs to process thermal and RGB images in real-time, enabling comprehensive detection of PV panel defects. <uri xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">https://github.com/TKing-Su/RTPV-YOLO</uri>.