Multi-modal PV defect dataset

We present a small novel multi-modal image dataset collected from a commercial Floating Photovoltaic (FPV) installation, combining thermal imaging with color information through an innovative channel-wise fusion approach.The dataset consists of 10 synchronized image pairs captured at 2.5m altitude under clear weather conditions, where the Red channel contains calibrated thermal data, while the Green and Blue channels incorporate the luminance (Y) and chrominance (U) components, respectively, derived from YUV-transformed RGB images.The dataset encompasses seven distinct classes of PV module conditions: Hotspots, Bird Droppings, Dirt Accumulation, Heated Junction Boxes, intact PV Modules, Residual Damage, and compound Hotspots caused by Bird Droppings or Dirt. This unique combination enables simultaneous analysis of thermal signatures and visual features within a single image structure, facilitating research in automated inspection and fault detection in floating solar installations.Each image is accompanied by detailed metadata including capture conditions, calibration parameters, and ground truth annotations for all seven classes. The dataset's structure allows for straightforward integration with existing computer vision frameworks while providing new opportunities for developing algorithms specifically tailored to FPV installation monitoring and maintenance. This resource addresses the growing need for specialized multi-modal datasets in the renewable energy sector and supports research in thermal-visual fusion applications for solar installation inspection.

本研究提出一款小型新型多模态图像数据集（multi-modal image dataset），采集自商用漂浮式光伏（Floating Photovoltaic, FPV）电站，通过创新性的通道级融合（channel-wise fusion）方法将热成像与彩色信息相结合。该数据集包含10对同步图像对，采集于晴朗天气条件下、2.5米高度处；其中红色通道承载经过校准的热成像数据，绿色通道与蓝色通道分别集成源自YUV色彩空间转换的RGB图像的亮度（Y）与色度（U）分量。数据集涵盖七类不同的光伏组件工况：热斑（Hotspots）、鸟粪（Bird Droppings）、积尘（Dirt Accumulation）、加热接线盒（Heated Junction Boxes）、完好光伏组件（intact PV Modules）、残余损伤（Residual Damage），以及由鸟粪或积尘引发的复合热斑（compound Hotspots）。这种独特的组合模式可在单幅图像结构内同时实现热特征与视觉特征的分析，为漂浮式太阳能电站的自动巡检与故障检测研究提供助力。每幅图像均附带详细元数据（metadata），包含采集条件、校准参数以及全部七类工况的真值标注（ground truth annotations）。该数据集的结构可便捷集成至现有计算机视觉框架（computer vision frameworks），同时为开发专门适配FPV电站监测与运维的算法提供了全新契机。本数据集响应了可再生能源领域对专业多模态数据集日益增长的需求，可为太阳能电站巡检中的热视觉融合（thermal-visual fusion）应用研究提供支撑。