Importance of Spatial and Spectral Data Reduction in the Detection of Internal Defects in Food Products

Abstract

Despite the importance of data reduction as part of the processing of reflection-based classifications, this study represents one of the first in which the effects of both spatial and spectral data reductions on classification accuracies are quantified. Furthermore, the effects of approaches to data reduction were quantified for two separate classification methods, linear discriminant analysis (LDA) and support vector machine (SVM). As the model dataset, reflection data were acquired using a hyperspectral camera in 230 spectral channels from 401 to 879 nm (spectral resolution of 2.1 nm) from field pea (Pisum sativum) samples with and without internal pea weevil (Bruchus pisorum) infestation. We deployed five levels of spatial data reduction (binning) and eight levels of spectral data reduction (40 datasets). Forward stepwise LDA was used to select and include only spectral channels contributing the most to the separation of pixels from non-infested and infested field peas. Classification accuracies obtained with LDA and SVM were based on the classification of independent validation datasets. Overall, SVMs had significantly higher classification accuracies than LDAs (P , 0.01). There was a negative association between pixel resolution and classification accuracy, while spectral binning equivalent to up to 98% data reduction had negligible effect on classification accuracies. This study supports the potential use of reflection-based technologies in the quality control of food products with internal defects, and it highlights that spatial and spectral data reductions can (1) improve classification accuracies, (2) vastly decrease computer constraints, and (3) reduce analytical concerns associated with classifications of large and high-dimensional datasets.

Using hyperspectral imaging to determine germination of native Australian plant seeds

“…In addition, we conducted 1 Â 2 spectral binning (decreased the spectral resolution from 2.1 to 4.2 nm), which resulted in 110 spectral bands being included in the analysis. Spectral binning was deployed, as it has been shown to increase classification accuracy [23,40]. Similar previously published studies [24,40], a radiometric filter was applied to exclude background, so that a pixel was only included, if the reflectance value of Acacia and Banksia seed coat at 660 nm (R660) met the following criterion: 0:050 < R660 < 0:250…”

“…Several studies have demonstrated the potential of reflectancebased spectroscopy methods in studies of plant seeds, including detection of internal infestations by weevils (Bruchus pisorum) in dry field peas (Pisum sativum) [23,24], classification of near isogenic maize lines (Zea mays) [25], ageing of cabbage seeds [26], classification of near isogenic maize lines (Z. mays) [27], differentiation between black walnut (Juglans nigra) shell and pulp [28], sorting of lettuce (Lactuca sativa) seeds [29], and viability of horticultural seeds [26,30,31]. These spectroscopy studies are based on the fundamental assumption that reflectance data acquired from the seed coat provides indicative information about the quality/ germination of the given seed.…”

Increased Susceptibility to Aphids of Flowering Wheat Plants Exposed to Low Temperatures

“…In addition, we conducted 2 by 2 spectral binning (increased the spectral resolution from 2.1 to 4.2 nm) and omitted the first and last 10 spectral bands (these spectral bands are typically associated with proportionally higher levels of stochasticity or noise), which resulted in 111 spectral bands from 411-871 nm being included in the analysis. The spectral and spatial binning was conducted to increase classification accuracy (Nansen et al 2013b, Zhang et al 2015 and to reduce the risk of model overfitting (Kemsley 1996, Defernez and Kemsley 1997, Faber and Rajko 2007.…”

Remote Sensing and Reflectance Profiling in Entomology

“…Imaging software engineers have developed numerous approaches to fragment pixels (1,55) and to artificially increase the spatial resolution of one imaging source through the use of a second type of imagery as part of a process called image fusion (35,128). Despite these classification solutions, it is generally recommended that the spatial resolution be high enough to avoid mixed pixels and then averaged as part of spatial binning (reducing the spatial and spectral resolutions of hyperspectral imaging data) of input data (42,65,158). However, acquisition of reflectance data at a spatial resolution high enough to avoid mixed pixels becomes a serious challenge when remote sensing data are acquired from small plants, such as newly established crop plants, and from cereals or other plants with partially vertical leaves (and therefore have a small footprint when imaged from above).…”

“…However, there are comparative studies in which both multispectral and hyperspectral systems enabled investigators to accurately classify and detect biotic stressors in crops (149). Furthermore, experimental benchtop remote sensing studies suggest that both spatial and spectral binning may partially improve the ability to detect stress responses (94,158). Possible constraints associated with acquiring remote sensing data at a high spectral resolution include increased equipment costs, data acquisition at lower spatial resolution (see below), data storage constraints, and, in airborne remote sensing, a lack of powerful airborne devices with higher pay load.…”