A Proximal Near-Infrared Hyperspectral Imaging Dataset of Cryptogamic Communities

"Cryptogams_NIR_HSI_Dataset" consists of close-range NIR-HSI images of naturally occurring cryptogams such as chlorolichen, cyanolichen, bryophyte (moss), and bark. The samples were collected from a natural environment in Hamilton, New Zealand, and then imaged using the help of Specim FX17e push-broom close-range HSI camera capturing the spectrum between 900-1700 nm with 224 spectral bands in controlled laboratory conditions. In total 39 hyperspectral cubes were captured and manually annotated using in-house developed HAPPy (Hyperspectral Application Platform in Python) annotation tool. Manual annotations were sparse and dense pixel-wise labels depending on the human annotator’s confidence; the value of an "ignore" label (255) was assigned to uncertain and unlabelled specimen regions during preprocessing. Out of 39 samples present in the dataset, 29 were labeled while 10 were unlabeled. Raw data were transformed into relative reflectance values and SNV normalized data and stored in unified HDF5 (.h5) format that includes Raw cubes, SNV normalized cubes (370 × 520 × 224), semantic masks (370 × 520), and wavelength vectors (224 bands). The presented dataset was collected and curated by the University of Waikato (WaI2M: Waikato Instrumentation and Measurement Research Group, & Hyperspectral Imaging Group), and it can be used in studies related to hyperspectral image processing, spectral analysis, cryptogamic detection, and semantic segmentation using machine learning/deep learning approaches. The dataset was used in the following accepted conference paper: "Near-infrared hyperspectral imaging and deep learning for semantic segmentation of cryptogams with sparse annotations" (Paper ID: 1571310702) to be presented at the IEEE 12th International Conference on Smart Instrumentation, Measurement and Applications (ICSIMA 2026). This paper discusses the preparation of the dataset, the strategy for sparse annotation, and the deep learning techniques for semantic segmentation of cryptogamic classes. In case of using the dataset in your research publication, please acknowledge the use of the dataset and its corresponding paper. @INPROCEEDINGS{1571310702, author={Faisal, Shah and Ooi, Melanie Po-Leen and Kuang, Ye Chow and Abeysekera, Sanush K. and Thawdar, Yaminn}, booktitle={2026 IEEE 12th International Conference on Smart Instrumentation, Measurement and Applications (ICSIMA)}, title={Near-Infrared Hyperspectral Imaging and Deep Learning for Semantic Segmentation of Cryptogams With Sparse Annotations}, year={2026}, volume={}, number={}, pages={}, address={Kuching, Sarawak, Malaysia}, note= {Accepted paper, Paper ID: 1571310702} }