A Time Series Dataset of NIR Spectra and RGB and NIR-HSI Images of the Barley Germination Process
Ole-Christian Galbo Engstrøm [email protected], Erik Schou Dreier [email protected], Birthe Møller Jespersen [email protected], Kim Steenstrup Pedersen [email protected] · 2025
Read A Time Series Dataset of NIR Spectra and RGB and NIR-HSI Images of the Barley Germination Process by Ole-Christian Galbo Engstrøm [email protected], Erik Schou Dreier [email protected], Birthe Møller Jespersen [email protected], Kim Steenstrup Pedersen [email protected] (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2504-16658.txt.
Introduction
Ole-Christian Engstrøm and three co-authors release an open data set on sprouting barley in a 2025 paper. Malting depends on even germination, so tracking how each kernel sprouts over time is valuable. The data set follows 2,242 single kernels through the process.
Every kernel was imaged before wetting and then every 24 hours for five days. The images come in three kinds: normal color, near-infrared spectra, and near-infrared hyperspectral images, each with masks that outline the kernels. Each kernel is labeled as sprouted or not at every step. Because the background is black filter paper, simple threshold methods can cut the kernels out. The set supports work on predicting sprouting time from images or spectra.
Catalog record
| CREATOR | Ole-Christian Galbo Engstrøm [email protected], Erik Schou Dreier [email protected], Birthe Møller Jespersen [email protected], Kim Steenstrup Pedersen [email protected] |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 34 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-16658
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
A Time Series Dataset of NIR Spectra and RGB and NIR-HSI Images of the Barley Germination Process Ole-Christian Galbo Engstrøm [email protected], Erik Schou Dreier [email protected], Birthe Møller Jespersen [email protected], Kim Steenstrup Pedersen [email protected] Abstract 1 Introduction We provide an open-source dataset of RGB and NIR-HSI (near-infrared hyperspectral imaging) images with associated segmentation masks and NIR spectra of 2242 individual malting barley kernels. We imaged every kernel pre-exposure to moisture and every 24 hours after exposure to moisture for five consecutive days. Every barley kernel was labeled as germinated or not germinated during each image acquisition. The barley kernels were imaged with black filter paper as the background, facilitating straight-forward intensity threshold-based segmentation, e.g., by Otsu’s method ( Otsu 1979 ). This dataset facilitates time series analysis of germination time for barley kernels using either RGB image analysis, NIR spectral analysis, NIR-HSI analysis, or a combination hereof. This report contains an overview of the dataset in Section 2 , a description of the image and spectrum acquisition in Section 3 , and a description of the barley kernel germination process in Section 4 . These sections contain sufficient information for modeling purposes for understanding the conception of the dataset. We kindly thank our sponsors in Section 5 . In Appendix A , we give a complete and detailed description of the image acquisition process and the methods used to identify and track individual kernels throughout several imaging sessions. 2 Dataset overview The dataset consists of images of 90 Petri dishes, most with 25 barley kernels. A polymer grid is placed inside each Petri dish so that each grid cell is occupied by exactly one barley kernel. During acquisition, the Petri dishes were imaged simultaneously by an RGB camera and a NIR-HSI camera, yielding RGB and NIR-HSI images of the same barley kernels under the same conditions. Each Petri dish was imaged before exposure to moisture and after exposure to moisture every 24 hours for the next five days. Table 1 shows an example of tracking a single barley kernel throughout the six imaging sessions. Figure 1 shows an example of an RGB and NIR-HSI pair of images of the same Petri dish taken simultaneously. This dataset consists of four different samples of malting barley: two of the variety Prospect and two of the variety Laureate. Table 2 shows the number of kernels from each variety. Table 3 and Figure 2 show the number of kernels classified as germinated after each day. Pre moisture Day 1 Day 2 Day 3 Day 4 Day 5 RGB NIR-HSI Table 1 : A single barley kernel imaged before exposure to moisture and every 24 hours after exposure to moisture. A sprout appears on Day 2, meaning that the reference value for germination time for this specific kernel is two days. ((a)) RGB. ((b)) NIR-HSI. Figure 1 : A Petri dish imaged at day 1. The barley kernel inside the top left square is the one shown in Table 1 . Barley Variety Prospect_0 Prospect_1 Laureate_0 Laureate_1 Total No. kernels 624 394 624 600 2242 No. Petri dishes 25 16 25 24 90 Table 2 : Number of barley kernels and Petri dishes for each of the four barley varieties. Germination Day Barley Variety 1 2 3 4 5 Any Prospect_0 4 29 23 15 12 83 Prospect_1 1 0 1 0 3 5 Laureate_0 13 21 25 15 17 91 Laureate_1 9 43 57 58 35 202 All 28 93 106 88 67 382 Table 3 : Number of barley kernels that germinated each day out of those shown in Table 2 . No kernels were germinated prior to exposure to moisture. ((a)) Prospect_0. ((b)) Prospect_1. ((c)) Laureate_0 ((d)) Laureate_1 ((e)) All varieties. Figure 2 : Histograms showing how the number of germinated kernels evolves from day to day. 2.1 Dataset availability We release the dataset and license it under CC BY-NC 4.0. It is available for download at The University of Copenhagen’s Electronic Research Data Archive . We provide the germination labels and images of extracted kernels, as shown in Table 1 . We also provide segmentation masks for all images to distinguish the barley kernels from the background. We also provide an NIR absorbance spectrum computed as an average of the pseudo absorbance inside the segmentation mask for each NIR-HSI image. Should dataset users be interested in applying their grain kernel detection or segmentation methods, we also provide the dataset of entire Petri dish images and the coordinates of the 36 grid corners, as shown in Figure 1 . Initially, we had envisioned 100 Petri dishes with 25 barley kernels each, totaling 2500 kernels. However, due to mold infections and human error, we lost ten Petri dishes and a few individual kernels from the remaining 90 Petri dishes. The retained dataset contains 2242 kernels. Every barley kernel in the retained dataset is imaged all six times with both cameras. 3 Image and spectrum acquisition 3.1 Imaging setup Figure 3 illustrates the imaging setup, which consists of a conveyor band moving a Petri dish with grain kernels underneath two line scan cameras, an RGB camera, and a NIR-HSI camera. On both ends of the setup are three 20-watt halogen light bulbs, ensuring bright illumination throughout the visual and NIR spectrum. During image acquisition, a Petri dish is placed on a plate carrying four quadratic chessboards for pixel size standardization and two windings of white PTFE (polytetrafluoroethylene) foil serving as white references. Dark references were acquired for the NIR-HSI camera by taking images with the shutter closed. For the RGB camera, no dark reference was necessary. See Appendix A for further details on the image standardization process. The RGB camera is a Basler Ace 2 Pro, while the NIR-HSI camera is a Specim FX17, with 224 uniformly distributed wavelength channels in the 900-1700nm spectrum. The RGB camera uses the standard 8 bits for each color channel, and the images are therefore stored as uint8 arrays. However, The NIR-HSI camera uses 12 bits for each wavelength channel, requiring 1.5 bytes. We store the NIR-HSI images in the mono12p format to facilitate optimal storage. This format packs two 12-bit values sequentially in three bytes (24 bits). Along with the dataset, we provide functions to convert both ways between mono12p and uint16, allowing easy analysis of uint16 images while reducing storage requirements by saving them as mono12p images. The functions are written using the Python library NumPy ( Harris et al. 2020 ). 3.2 Image acquisition and single kernel tracking Each Petri dish was imaged simultaneously with an RGB and an HSI camera during image acquisition. The polymer grid inside the Petri dish enables straightforward extraction of individual kernels from entire Petri dish images. To extract a single barley kernel, we rely on the polygon defined by the coordinates of the four surrounding grid points. To extract a rectangular image, we compute the circumscribed rectangle of the polygon, making sure to mask out (set to zero) any part of the rectangle outside the polygon. This process yields single kernel images, as shown in Table 1 . The ArUco codes are fixed on the grids, which are, in turn, fixed on the Petri dishes. Thus, based on the location