HyperspectralBlueberries: a dataset of hyperspectral reflectance images of normal and defective blueberries

The HyperspectralBluberries dataset consists of hyperspectral datacubes, which were acquired by an in-house assembled benchtop line scanning system, from 420 blueberries of two categories, including 210 sound fruit and 210 samples with various defects. The fruit samples were hand-picked from a commercial orchard. Each scanning event, which was done for an array of 42 samples, yields two files in image formats .bil (band-interleaved-by-line) and .hdr (header), which store the hyperspectral raw data and associated metadata, respectively, and are both necessary for loading hyperspectral data for processing.  In addition to sample scanning, a white reference was also scanned, which can be used for standardizing spectral responses. As a result, there are 22 files in the dataset, totaling about 25 GB in file size. The sample file names are descriptive, indicating the blueberry category and number information. The dataset was used for developing machine learning models for differentiating between normal and defective blueberries, achieving an overall accuracy of 96.6%. Software programs for the modeling work are publicly available at: https://github.com/vicdxxx/Blueberry-Defect-Detection-by-Hyperspectral-Imaging.

Details about the dataset curation and modeling experiments are described in the journal article: Deng, B., Lu, Y., Stafne, E. (2024). Fusing Spectral and Spatial Features of Hyperspectral Reflectance Imagery for Differentiating between Normal and Defective Blueberries. Smart Agricultural Technology. https://doi.org/10.1016/j.atech.2024.100473. If you use the dataset in published research, please consider citing the dataset or the journal article. Hopefully, you find the dataset useful.