SWIR hyperspectral data cubes for plastics detection in the environment

The dataset contains 9 hyperspectral cubes related to paper "Attention‑Gated U‑Net for Robust Cross‑Domain Plastic Waste Segmentation using UAV Based Hyperspectral SWIR Sensor" by S. Bouchelaghem, M. Balsi, M. Moroni, submitted in November 2025 to Remote Sensing Applications: Society and Environment. Corresponding author: [e-mail] Data were acquired using a push-broom SWIR camer mounted on a drone, in controlled natural environments, where sorted plastics objects were placed on the ground. The files contain hyperspectral cubes organized as 81 layers corresponding to wavelengths from 900 to 1700 nm, sampled every 10nm in the folder "drone_cube". Layers from 900 to 930 are dummy (filled with zeros) because they were not actually acquired. For each cube, manually-drawn masks are provided, for labelling according to plastics polymer or other material in the folder "masks". Additional folder "Training_dataset2" include the training used for the deep learning model.本数据集包含9个高光谱立方体（hyperspectral cubes），相关研究论文为S. Bouchelaghem、M. Balsi、M. Moroni于2025年11月提交至《Remote Sensing Applications: Society and Environment》的《Attention‑Gated U‑Net for Robust Cross‑Domain Plastic Waste Segmentation using UAV Based Hyperspectral SWIR Sensor》，通讯作者邮箱为[e-mail]。 数据通过搭载于无人机的推扫式短波红外（SWIR）相机采集，采集场景为受控自然环境，地面放置了分类后的塑料物件。 在"drone_cube"文件夹内，高光谱立方体文件包含81个波段层，对应900 nm至1700 nm的波长范围，采样间隔为10 nm。其中900 nm至930 nm的波段层为无效数据（填充为0），因该波段范围未实际采集。 在"masks"文件夹中，为每个高光谱立方体提供了人工绘制的掩码，用于按照塑料聚合物或其他材料类别进行标注。 额外文件夹"Training_dataset2"包含用于该深度学习模型训练的训练数据集。