Evaluation of Strawberry Quality using Near-Infrared Hyperspectral Imaging(NIR-HSI)
概要
報告番号
※
第
主
号
論
Evaluation
of
文
の
Strawberry
要
Quality
旨
using
Near-Infrared
Hyperspectral Imaging (NIR-HSI)
論文題目
(近赤外ハイパースペクトラルイメージングによるイチゴの
品質評価)
氏
名
関
隼人
論 文 内 容 の 要 旨
Developing non-destructive quality assessment techniques in agriculture
enhances our understanding of the connection between human senses and farm
products, aiding value determination, breeding, and crop improvement. Despite
the development of sugar content sensors, destructive methods still largely
dominate. The rise of non-destructive methods could provide accurate sensory
evaluation estimates and assess ingredient distribution within farm products,
improving quality control and value. In this study, we employed Near-Infrared
Spectroscopy (NIR), a non-destructive evaluation technique that irradiates
near-infrared light on a sample. Spectroscopy of the transmitted and reflected
light discloses molecular vibration characteristics, such as O-H, C-H, and N-H.
However, overlapping these absorption bands necessitates using a calibration
model, created through analytical methods like chemometrics. Although NIR
focuses on a point, the superior transmission properties of near-infrared light
facilitate non-destructive measurements. Moreover, Near-Infrared Hyperspectral
Imaging (NIR-HSI) expands upon NIR by providing a spatial measurement
technique. With its planar spectroscopy characteristic, NIR-HIS is potentially
suitable for evaluating agricultural products exhibiting varied spatial qualities.
In this study, we put forward a novel pretreatment in
quality evaluation
method for strawberry fruit, based on the region of interest (ROI) in the flesh
part and the spectral feature differences between the flesh and achene. Other
areas of focus include correcting hyperspectral data according to fruit shape,
providing a 3D display, and evaluating the entire strawberry fruit surface - all of
which have been overlooked in prior strawberry quality evaluation research
using NIR-HSI.
Our approach enables the visualization of the spatial distribution of sugar
content in white strawberry fruit flesh, employing NIR-HSI (913–2166 nm). We
scrutinized NIR-HSI data sourced from 180 samples of “Tochigi iW1 go” white
strawberries. Identifying pixels corresponding to the flesh and achene on the
strawberry surface necessitated principal component analysis (PCA) and image
processing (PCA imaging), following the data's smoothing and standard normal
variate (SNV) pretreatment. We applied Explanatory Partial Least Squares
Regression (PLSR) analysis to develop a model to predict Brix reference values.
This model, constructed from raw spectra extracted from the flesh ROI, exhibited
high predictive accuracy, with root mean square error of prediction (RMSEP) and
coefficient of determination of prediction (R 2 p) values of 0.576 and 0.841,
respectively, and with relatively low number of PLS factors. The Brix heatmap
images and violin plots for each sample revealed characteristic sugar content
distribution within the strawberries' flesh.
This study also incorporated the use of a line-scan hyperspectral camera and a
laser displacement meter to capture the hyperspectral image and shape of
strawberries of the 190 samples of “Tochigi i37 go”. Given the influence of the
strawberry shape on the recorded hyperspectral image, we devised a method to
correct the hyperspectral data. Corrections accounted for the distance through
the light attenuation of the inverse square of the distance and the angle using
the
Lambert
Cosine
method.
Furthermore,
this
correction
changed
the
characteristics of the spectra. In addition, the flesh was extracted by PCA
imaging. We conducted PLSR and optimal model search to validate the shape
correction and spectral preprocessing. As a result, the highest prediction
accuracy model (R2:0.813, RMSEP:0.687) was constructed after height and angle
correction and smoothing spectral preprocessing. Utilizing this PLSR model, we
created sugar content estimation mapping images for each condition, which were
then compared using the Map score index, developed to assess the quality of the
images. The model yielding the highest Map score of 52.5 combined height
correction and smoothing spectral processing (R2:0.791, RMSEP:0.727). It was
also discerned that second-derivative processing added noise to the images. By
combining the mapping image and shape data, we developed a synthesized 3D
sugar content image to illustrate the shape and sugar content distribution. In
addition, we proposed a Strawberry deviation (T-score) to assess the sugar
content value as NIR-HSI estimated.
To visualize sugar content distribution within the strawberry flesh section, we
constructed a system where strawberries were affixed to a turntable, and
rotation scan measurements were conducted using a hyperspectral camera and a
laser displacement meter. We obtained the 130 samples of “Tochigi i37 go” shape
and hyperspectral data. We studied shape correction and spectral preprocessing
from these measurements, constructed a PLS model and mapping, and found that
the model with height correction and smoothing preprocessing demonstrated good
prediction performance (R 2 :0.892, RMSEP:0.503) and imaging result. As the
measured data represented one full rotation of the fruit, we developed a 3D
model for visualizing and understanding the sugar content of strawberry flesh by
incorporating an angle and integrating the shape and sugar content mapping
results in 3D.
In conclusion, this study successfully visualized the sugar content distribution
of white strawberry flesh using NIR-HSI. It utilized PCA imaging to separate the
fruit surface from the flesh, facilitating sugar content imaging and distribution
evaluation. Additionally, it introduced a method for 3D sugar content imaging of
strawberries, incorporating shape measurements into the NIR-HSI method. The
study investigated the corrective effects of shape adjustment (height & angle) on
hyperspectral data and created a 3D representation of shape and sugar content
distribution. We proposed an evaluation method for the NIR-HSI mapping results
and introduced a rotation-NIR-HSI method, establishing a complete rotational
hyperspectral data measurement process for strawberries. We also developed a
3D model to visualize the sugar content distribution of strawberry flesh. This
research's findings can inform practical application methods for tasks such as
sorting, breeding, and variety improvement, and extending the shape and
objective variables to other internal quality factors (e.g., acidity, sugar-acid
ratio) will enable comprehensive quality evaluation.