Simulating single band multispectral imaging from hyperspectral imaging: A study into the application of single band visible to near-infrared multispectral imaging for determining table grape quality

Abstract

To be accepted by the market and consumers table grapes need to meet certain requirements in terms of physical and chemical quality parameters. Chemical quality parameters of importance include total soluble solids (TSS), total titratable acidity (TA) and pH (Herrera et al., 2003; Jayasena & Cameron, 2008). The standard reference methods used to determine these parameters are generally destructive, time-consuming and may require specialized equipment or complex sample preparation (Daniels et al., 2019; Fernández-Novales et al., 2019). Due to inherent intra-bunch and inter-bunch variability large samples are also required to ensure accurate representation of the vineyard. Alternative methods such as spectroscopy and spectral imaging may be the solution to providing fast, accurate and efficient evaluation of table grapes.

Spectroscopy methods including visible to near-infrared (VNIR) spectroscopy, Fourier-Transform near-infrared (FT-NIR) spectroscopy and short-wave infrared (SWIR) spectroscopy have been proposed and tested for the evaluation of fruit and vegetable quality. These methods are mostly laboratory-based and require highly sophisticated and expensive equipment. Being laboratory-based means that these methods still need samples to be removed from the vine i.e., are also destructive. Some in-field spectrometer options are available, but they remain expensive. Spectroscopy techniques are also generally limited to point measurements yielding one-dimensional data i.e., a spectral signature for the specific point. This makes it difficult to upscale measurements to a bunch-level investigation because the decision of how many points should be measured and what the distribution of these points across the bunch should be is not an easy one to make. Technologies such as hyperspectral imaging (HSI) and multispectral imaging (MSI) may assist in overcoming this challenge as they provide three-dimensional data comprised of the one-dimensional spectral and two-dimensional spatial data i.e., spectral signatures for each pixel in the image (Nicolaï et al., 2007).

HSI and MSI technologies are however not easily transferred to in-field applications due to the potential interference of ambient conditions (light conditions in particular) on the measured spectral data (Nicolaï et al., 2007). Some studies have successfully used HSI for in-field applications under specific conditions (Shao et al., 2021). Portable HSI sensors are still expensive and MSI options can provide a cost-effective alternative. MSI technologies have not been widely used in determining fruit quality and to our knowledge no studies on table grapes specifically exist. HSI provides a large amount of information, but MSI is generally limited to much fewer spectral bands or even single bands. This core difference may impact on the successful application of MSI technology for determining table grape quality parameters.