Potential applications of hyperspectral imaging for the determination of
total soluble solids, water content and firmness in mango

By
Sivakumar Servakaranpalayam. S.

Department of Bioresource Engineering
Macdonald campus, McGill University
Montreal, Quebec, Canada

A thesis submitted to the Graduate and Postdoctoral Office
in partial fulfillment of the requirements for
the degree of Master of Science

February 2006

©Sivakumar Servakaranpalayam. S., 2006


ABSTRACT
The application of hyperspectral imaging technique in the wavelength range of
400-1000 nm to estimate some of the maturity parameters of mangoes was investigated.
Mangoes with different quality levels were grouped using principle component analysis
(PCA). Feature wavelengths were identified to predict total soluble solids content, water
content and firmness using simple correlation, first derivative, partial least square (PLS)
regression analysis and measured values. Calibration models were developed using the
selected wavelengths from correlation coefficients, first derivative, partial least square
(PLS) regression analysis and corresponding maturity parameters employing artificial
neural network model to predict total soluble solids content, water content and firmness
of the fruit. Performance of the models was compared using the correlation coefficient (r)
values. Fruit firmness was predicted with high correlation coefficient (r=0.88) followed

by water content (r=0.81) and total soluble solids (r=0.78) using wavelengths selected
from simple correlation of first derivative data with the parameters and ANN model. The
results of the study demonstrated the scope for further research on maturity and quality
evaluation of fruits using hyperspectral imaging technique.

i


RÉSUMÉ
L’utilisation d’une technique d’imagerie hyperspectrale sur une gamme de
longueur d’ondes entre 400 et 1000 nm a été étudiée pour l’évaluation de certains des
paramètres de maturité de la mangue. Des mangues ayant des niveaux variés de qualité
ont été regroupées par une analyse composante principale.

Des longueurs d’ondes

particulières ont été identifiées pour la prédiction du contenu total de solides solubles, de
la teneur en eau et de la fermeté, et ce, en utilisant une simple corrélation, une dérivée de
premier ordre, une analyse de régression par la méthode des moindres carrés et des
données mesurées. Des modèles de calibration ont été développés à partir des longueurs
d’ondes issues des coefficients de corrélation, de la dérivée de premier ordre, de l’analyse
de régression par la méthode des moindres carrés et des caractéristiques de maturité
correspondantes et ce, en utilisant un réseau neuronal artificiel pour prédire le contenu
total de solides solubles, la teneur en eau et la fermeté des fruits. La performance des
modèles a été comparée en utilisant les coefficients de corrélation (r). La prédiction de
fermeté a obtenu un coefficient de corrélation élevé (r=0.88), suivi de la teneur en eau
(r=0.81) et le contenu total de solides solubles (r=0.78), en utilisant les longueurs d’ondes
issues d’une simple corrélation des données de la dérivée de premier ordre avec les
paramètres du modèle par réseau neuronal artificiel. Les résultats de l’étude ont défini
l’étendue de la recherche nécessaire pour l’évaluation de la maturité et de la qualité des

fruits par technique d’imagerie hyperspectrale.

ii


ACKNOWLEDGEMENTS
I have immense pleasure to express my deep sense of gratitude and
indebtedness to my thesis supervisor, Dr.G.S.Vijaya Raghavan, James McGill Professor
for suggesting this problem, his valuable suggestions and useful criticism, constant
encouragement and inspiration evinced throughout the course of this research and in
preparation of the thesis.
I wish to place on record my special gratitude to my thesis co-supervisor,
Dr.Ning Wang for her continuous and valuable guidance, support and motivation for
successful completion of the research work. My special thanks to Dr.Jun Qiao, for his
support and guidance extended to me during the course of this work. I am grateful to
Mr.Yvan Gariépy for his valuable help and support during this work. My special thanks
to M/s. Aliments IMEX Foods Inc., Montreal for having provided the mangoes for the
experiments.
I would like to acknowledge the Canadian International Development Agency
(CIDA) for their financial support under CIDA- UPCD TIER 1 project on “Consolidation
of Food Security in South India”. I am grateful to Tamil Nadu Agricultural University,
India for having permitted me to undergo this programme at McGill University, Canada.
I would like to express my gratitude to my Dean, Dr.R.Manian for his constant
encouragement and support during my stay at McGill University.
I have boundless pleasure in recording my profound thanks to Dr.Valérie Orsat
for her helpful guidance, constant inspiration and good wishes which led to the successful
completion of the project work and to Dr.Venkatesh Sosle for his encouragement and
support during the course of my work.
I would like to thank Ms. Susan Gregus, Ms. Trish Singleton and Ms. Abida
Subhan for their valuable help and support in all administrative aspects.

I express my sincere thanks to my lab mates in the “Instrumentation and Control
lab” Gamal ELMasry and Zhenfeng Li for their ready help, support and to my friends
Rajni Radja, Simona Nemes, Mariana Monroy and Laura Mesones for their constant
support and encouragement.
I express my deep sense of gratitude to my most affectionate and beloved
parents, wife and my sons Siddharth and Sri Aadhithya who sacrificed many things
during this period and made me to utilize this best opportunity in my life and to my
brother S.P.Balasubramanian and my sister B.Padmavathi for their encouragement,
timely and untiring help rendered during this period.
(S.S.SIVAKUMAR)

iii


CONTRIBUTION OF AUTHORS
This thesis research consists of three manuscripts authored by me. The
manuscripts were co-authored by Dr.G.S.V.Raghavan, my supervisor, Dr.Ning Wang,
my co-supervisor, Dr.Jun Qiao and Yvan Gariepy. Dr. Raghavan and Dr.Wang were my
direct advisory committee for formulation of the problem, designing the concepts,
methodology and execution of the thesis research and reviewed the thesis report.
Manuscript 1 is from the chapter 3 and it will be submitted for publication. Manuscript 2
from chapter 4 was submitted and accepted for oral presentation in the North Eastern
Agricultural and Biological Engineering (NABEC-2006), the 17th annual conference.
Manuscript 3 from chapter 5 has been accepted for oral presentation in Quality Detection
and Postharvest Handling of Fruits and Vegetables session of 2006 ASABE annual
meeting.

iv



TABLE OF CONTENTS
ABSTRACT

i

RÉSUMÉ

ii

ACKNOWLEDGEMENTS

iii

CONTRIBUTION OF AUTHORS

iv

TABLE OF CONTENTS

v

LIST OF TABLES

ix

LIST OF FIGURES

x

NOMENCLATURE


xi

I. GENERAL INTRODUCTION

1

1.1. Overview

1

1.2. Organization of thesis

3

1.3. Scope of the study

3

II. GENERAL OBJECTIVES

4

III. GENERAL LITERATURE REVIEW

5

3.1. Mango (Mangifera indica. L.)

5


3.2. Mango grading

6

3.3. Machine vision technology for agri-food industry

7

3.3.1. Back ground

7

3.3.2. Applications in agri-food industry

8

3.4. Visible / near infrared spectroscopy (VNIRS)

9

3.4.1. Back ground

9

3.4.2. Applications in agri-food industry

10

3.5. Hyperspectral imaging


12

3.5.1. Back ground

12

3.5.2. Theory of hyperspectral imaging technique

12

3.5.3. Advantages and disadvantages

13

3.6. Quality evaluation of fruits and vegetables using machine, VNIRS and
hyperspectral imaging

v

13


3.6.1. Machine vision

13

3.6.2. Near infrared spectroscopy

14


3.6.3. Hyperspectral imaging

18

3.6.4. Summary

20

3.7. Multivariate statistical methods for image and VNIRS analysis

21

3.7.1. Background

21

3.7.2. Typical multivariate statistical methods

22

3.7.2.1. Principle Component Analysis (PCA)

22

3.7.2.2. Partial Least Square (PLS)

23

3.7.2.3. Artificial Neural Network (ANN)


24

3.7.3. Wavelength selection

25

3.7.3.1. Principle Component Analysis (PCA)

25

3.7.3.2. Multiple Linear Regression (MLR)

26

3.7.4. Classification

26

3.7.4.1. Partial Least Square (PLS)

27

3.7.4.2. Artificial Neural Network (ANN)

27

3.8. Summary

28


REFERENCES

28

CONNECTING TEXT

37

IV. POTENTIAL USE OF HYPERSPECTRAL IMAGING FOR
SORTING MANGOES – A QUALITATIVE EVALUATION

38

4.1. Abstract

38

4.2. Introduction

38

4.3. Materials and methods

41

4.3.1. Fruit samples

41


4.3.2. The hyperspectral imaging system

41

4.3.3. Hyperspectral image acquisition

42

4.3.4. Data analysis

42

4.3.4.1. Principle Component Analysis

43

4.3.4.2. Feed-forward neural network

44

vi


4.4. Results and discussion

45

4.4.1. NIR spectra for mangoes

45


4.4.2. Principle component analysis

46

4.4.3. Classification by feed forward neural network

47

4.5. Conclusion

49

ACKNOWLEDGEMENT

49

REFERENCES

49

CONNECTING TEXT

51

V. DETECTING MATURITY PARAMETERS OF MANGO USING
HYPERSPECTRAL IMAGING TECHNIQUE

52


5.1. Abstract

52

5.2. Introduction

52

5.3. Materials and methods

54

5.3.1. Mango samples

54

5.3.2. Hyperspectral image data collection

55

5.3.2.1. Image processing algorithm

55

5.3.2.2. Image acquisition

56

5.3.3. Chemical/mechanical attributes


57

5.3.3.1. Total soluble solids

57

5.3.3.2. Water content

57

5.3.3.3. Firmness

57

5.3.4. Image calibration and data analysis

58

5.3.4.1. Spectral data

58

5.3.4.2. Image calibration

59

5.3.4.3. Spectral analysis and feature wavelength selection

59


5.3.4.4. Feed-forward neural network

60

5.4. Results and discussion

60

5.4.1. Results from the chemical analysis

60

5.4.2. VIS/NIR calibration results

61

5.4.3. Selection of feature wavelengths

62
vii


5.4.3.1. Model 1: Simple correlation model using the original data

62

5.4.3.2. Model 2: Correlation model using the first derivative data

63


5.4.3.3. Model 3: PLS calibration model

65

5.4.4. Feed-forward neural network model

65

5.4.5. Hyperspectral image spectra analysis

67

5.4.5.1 Total soluble solids content prediction

67

5.4.5.2. Prediction of water content

68

5.4.5.3 Firmness prediction

69

5.5. Conclusions

71

ACKNOWLEDGEMENT


71

REFERENCES

71

VI. GENERAL CONCLUSIONS

73

ANNEXURES

75

viii


LIST OF TABLES
Table 4.1

Results of FFN network model for 500-1000 nm spectra

48

Table 4.2

Results of FFN network model for 700-1000 nm spectra

48


Table 5.1

Mean values of chemical/mechanical attributes of the
samples

61

Table 5.2

Selected wavelengths from correlation analysis of the
original data

63

Table 5.3.

Selected wavelengths from correlation analysis of the first
derivative data

64

Table 5.4.

Selected wavelengths from PLS regression analysis

65

Table 5.5.

Results of ANN model from the original and first

derivative data

66

Table 5.6.

Results of ANN model from the PLS analysis

66

Table 5.7.

Prediction of total soluble solids content

67

Table 5.8.

Prediction of water content

68

Table 5.9.

Prediction of firmness

69

ix



LIST OF FIGURES
Figure 4.1

Hyperspectral image acquisition system

41

Figure 4.2

Spectral curves for mangoes at different duration

45

Figure 4.3

Principle component analysis for 500-1000 nm spectra

46

Figure 4.4

Principle component analysis for 700-1000 nm spectra

47

Figure 5.1

Flowchart of algorithm for the prediction of maturity
parameters


55

Figure 5.2

Hyperspectral image acquisition system

56

Figure 5.3

Firmness measurement using Instron testing system

58

Figure 5.4

Mean reflectance spectrum for mangoes

61

Figure 5.5

Correlation of spectral data with the chemical attributes

62

Figure 5.6

Correlation of first derivative spectral data with chemical

attributes

63

Figure 5.7

Prediction results for total soluble solids content using feedforward neural network with inputs of wavelengths from first
derivative data and measured values

68

Figure 5.8

Prediction results for water content using feed-forward
neural network with inputs of wavelengths from first
derivative data and measured values

69

Figure 5.9

Prediction results for firmness using feed-forward neural
network with inputs of wavelengths from first derivative data
and measured values

70

x



NOMENCLATURE
°brix

-

Degree brix

°C

-

Degree Celsius

ANN

-

Artificial Neural Network

ATR-FTIR

-

Attenuated
Total
Transform InfraRed

CCD

-


Charge-Coupled Device

CCIR

-

Consultative Committee for International Radio

CMOS

-

Complementary Metal-Oxide Semiconductors

CP

-

Crude Protein

DC

-

Direct Current

DM

-


Dry Matter

ENVI

-

Environment for Visualizing Images

FAO

-

Food and Agriculture Organization

FFN

-

Feed-Forward Neural

FLD

-

Fishers’s Linear Discriminant

GUI

-


Graphical User Interface

IR

-

InfraRed

MATLAB

-

MATrix LABoratory

MIR

-

Mid-InfraRed

MLR

-

Multiple Linear Regression

MPLS

-


Modified Partial Least Square

MRI

-

Magnetic Resonance Imaging

NIR

-

Near InfraRed

NIRS

-

Near InfraRed Spectroscopy

NTSC

-

National Television System Committee

PAL

-


Phase Alternating Line

PCR

-

Principle Component Regression

PLS

-

Partial Least Square

PLSR

-

Partial Least Square Regression

ROI

-

Region of Interest

SEC

-


Standard Error of Calibration

xi

Reflectance-

Fourier


SEP

-

Standard Error of Prediction

SSC

-

Soluble Solids Content

S-VHS

-

Super Vertical Helix Scan

TSS


-

Total Soluble Solids

US

-

United States

USDA

-

United States Department of Agriculture

UV

-

Ultra Violet

VIS

-

Visible

VNIRS


-

Visible/Near InfraRed Spectroscopy

λ

-

Wavelength

xii


I. GENERAL INTRODUCTION
1.1. Overview
Mango (Mangifera indica. L.) is an important commercial fruit crop throughout
the world particularly in South East Asian countries, e.g. India, Malaysia, Indonesia,
Sri Lanka, Thailand and also in African countries, e.g. Egypt, South Africa, Islands of
West Indies. North America is the major importers, which accounts for 42% of global
fresh mango imports. Asia continues to be the leader in mangoes production with 76.9%
of total world mango production and Mexico is the largest exporter of mango with 41%
share in the world market. North American imports increased around 40% when
compared between the year 1996 and 2000 according to the FAO statistics database
(Galán Saúco, 2004).
Post harvest handling of mango is an important area which needs to be addressed
when the produce is exported to distant markets. As a key factor for increasing the shelf
life of the fruit under different storage conditions, correct stage of maturity is defined as
the stage of the fruit at which the produce maintains its natural taste and flavour.
In general, the maturity of fruits is determined by date, skin colour, taste, flesh
firmness, soluble solids content, stem-to-fruit removal force, respiration rate and ethylene

production rate. Maturity detection based only on skin colour and flesh firmness often
leads to erroneous predictions. Hence, maturity detection with information on fruits
chemical and soluble solids changes, respiration, and ethylene production rate will be
more scientific and appropriate when it goes for export markets. Currently most of the
methods of estimating chemical component changes, respiration and ethylene changes are
destructive. From the earlier research studies, it was observed that reducing sugar content

1


was constant from immature to full ripened stage and same is the case with the acid
content. But the sucrose content increased during the ripening process. As the maturity
progresses, the water content in the fruit decreases with increase in total sugar content
and the fruit becomes softer. Hence nondestructive methods to determine the total soluble
solids content, water content and firmness would be most appropriate in quality
evaluation based on maturity of the fruit.
Machine vision, spectroscopy and hyperspectral imaging or imaging spectroscopy
are potential fields of research utilizing the optical properties of objects to study the
internal properties. Applications of these techniques in quality and safety evaluation,
classification and sorting of food materials, fruits and vegetables is becoming a vital
scientific approach for ensuring quality products in the market. Recent research in
machine vision is being concentrated more towards the ultraviolet (UV), near infrared
(NIR) and infrared (IR) ranges of light spectrum which are invisible to human
(Chen et al., 2002). The information gathered from the objects at these invisible regions
are useful in determining the presence of diseases, defects and damages, pre and post
harvest maturity and internal quality of the objects.
Considerable amount of research works have been carried out on visible/NIRS for
determining internal quality of fruits and vegetables, detection of defects and diseases.
A very small area or a point is being used in spectroscopy for detecting the internal
properties of the materials. Hyperspectral imaging or imaging spectroscopy is gaining the

attention of researchers in the field of food research in recent days because of its
advantages over spectroscopy. It uses the entire area of the object to evaluate the internal
properties which has the major advantage of predicting very minute details with high
precision. With the recent advances in the field of computers, digital cameras and
2


computer software technologies which can perform image capturing and image
processing operations with very high speed and precision, the adoption of hyperspectral
imaging technique becomes more feasible for better evaluation of quality parameters.
1.2. Organization of thesis
This thesis report has five chapters. The first chapter explains the present scenario
of mango industry in the world, existing problems in quality detection, grading and
sorting and scope of this research. The second chapter illustrates the general goal with
emphasis on specific objectives of the study. Review of earlier work in this area of
research is presented in Chapter 3. Qualitative evaluation and selection of feature
wavelengths using hyperspectral imaging is reported in Chapter 4.
Chapter 5 describes development of models for the prediction of total soluble
solids content, water content and firmness for mangoes using hyperspectral imaging
technique. Chapter 6 summarizes the conclusions derived from this research.
1.3. Scope of the study
The present study aims at exploring the potential use of hyperspectral imaging for
determining the maturity parameters of mangoes. It includes identification of feature
wavelengths in visible/NIR spectra for estimating total soluble solids, water content and
firmness of mangoes. This work also focuses on development of calibration models for
predicting the maturity parameters from the spectral information.
The results from this study could be used in online sorting of mangoes for export
and consumer markets based on their total soluble solids content, water content and
firmness.
The results obtained from this research study are specific to the conditions stated.


3


II. GENERAL OBJECTIVES
The objectives of this study were,
1. To assess the potential of hyperspectral imaging system to classify different
grades of mangoes
2. To select the feature wavelengths for determining the total soluble solids content,
water content and firmness
3. To develop calibration models for the prediction of total soluble solids content,
water content and firmness
4. To adopt the hyperspectral imaging technique to quantify water content and total
soluble solids content of the fruit

Experiments were carried out in the Department of Bioresource Engineering,
McGill University, Canada from September 2004 to December 2005.

4


III. GENERAL LITERATURE REVIEW
3.1. Mango (Mangifera indica. L.)
Mango (Mangifera indica. L.) is a tropical and subtropical fruit originated in the
indo-Burma region and has been cultivated for over 4000 years in India (Nakasone and
Paull, 1998). Mangoes are available in fresh form throughout the world including North
America, Japan and Europe apart from the mango producing countries (Mukherjee,
1997). Mango is consumed as a fruit and as a processed product like nectar, squash, pulp,
juice, jam, jellies, leather and canned slices.
Main constituents of mango fruit are water, carbohydrates, acids, protein, fats,

minerals, pigments, tannins, vitamins and flavour. For almost all varieties of mango
grown throughout the world, the moisture content is 79.1 to 86.1%, citric acid is 0.12 to
1.44%, oil content is 0.03 to 0.92%, protein is 0.22 to 1.12%, total soluble solids is 11.5
to 25.0% and ash 0.26 to 1.16% (Singh, 1960). The ripe mango fruit is a good source of
vitamins A and C (Gandhi, 1955).
Mango is harvested after attaining its true physiological maturity stage. For export
markets, it is harvested as green fruit having completed its full size and maturity.
Harvesting is usually by hand picking or by mango pickers. Mango is a climacteric fruit
which continues to ripen after harvested from the tree. For mangoes, as the maturity
progresses, the skin colour changes, flesh firmness increases, soluble solids content
increases and the taste of the fruit changes. The maturity indices for mango are, change in
shape of the fruit, change in skin colour from dark-green to light-green to yellow, and
change in flesh colour. The quality indices are shape, size, skin colour, flesh firmness,
injury, defects and changes due to ripening such as increased sweetness, decreased

5


acidity and increased aroma volatiles and carotenoids. The flavour quality constituted by
sweetness, textural quality, sourness and aroma varies among different varieties.
After harvesting, the fruits are washed to remove sap from surface to prevent sap
burn and treated with hot water, a fungicide or chlorine at 52°C for 5 minutes to control
anthracnose. The shelf life of mango at ambient conditions is 7-14 days to fully ripe
which also varies with the variety, maturity at harvest and presence of injuries. The
recommended storage temperature for a fully ripe fruit is 8-10°C.
3.2. Mango grading
Mango grading is an important postharvest operation. If the immature fruits are
sent to the market along with matured ones, they often exhibit erotic ripening behaviour
and develop off-flavours and aroma which spoil other fruits and reduce the commercial
value. In cases where over-matured fruits are mixed with mature/immature fruits, the

ethylene from the matured fruit enhances the ripening which in turn reduces the shelf life
of the fruits. Hence, correct determination of maturity levels and sorting based on
maturity becomes vital.
Basically the Mango grade standards are defined based on shape, size, colour and
levels of defects/damages. The United States Department of Agriculture (USDA) has
fixed standards for all grades of mango which includes definition for clean, damage,
diameter, fairly well formed, injury, maturity, misshape, overripe, serious damage,
similar varietal characteristics, well formed and well trimmed mango for US consumer
market (USDA, 2006). The maturity in the standards is defined as the stage of the fruit
which ensures proper completion of the ripening process. The mango which yields to a

6


slight pressure is considered as disintegration and has crossed the commercial usage as
per the standards (USDA, 2006).
Presently the grading is mostly done manually based on the weight, variety, size,
shape, surface defects and stage of ripeness. The sorting is carried out by long-time
trained quality control inspectors only. The manual sorting or grading is a time
consuming, laborious, less efficient and inaccurate process. Hilton (1994) mentioned the
scope of automation for grading and packing to reduce the labour costs and to increase
the production.
Mechanical graders employing firmness sorter, size graders and weight sorters
are used in some countries. Hahn (2004) used a firmness sorter which could detect the
firmness with 95.7% accuracy and grade mangoes as hard, soft and very soft with 90%
accuracy at a speed of one fruit per second.
However when it comes to grading/sorting based on the internal properties of
mango it is still based on destructive tests which are difficult to apply for on-line or large
scale grading/sorting. In such applications the machine vision based grading techniques
employing the optical properties of fruit will be of more use for efficient and rapid

grading/sorting.
3.3. Machine vision technology for agri-food industry
3.3.1. Back ground
Machine vision is an advanced technology to “see” objects with an assistance of
computers. It has been used in many applications in agriculture such as remote sensing,
precision farming, pre- and postharvest product quality and safety detection, and sorting.
The use of machine vision is increasing rapidly because of the advantage of providing the

7


numerical attributes of the object in addition to the information about the colour, size,
shape and textural attributes (Chen et al., 2002).
The quality assessment by machine vision has a wide range of applications in
agriculture to detect presence of diseases, defects, ripeness, maturity, and other quality
attributes in fruits and vegetables, grain and animal products etc. The main advantages of
machine vision technology are that the results or information are accurate and consistent
and also nondestructive. This will enhance good quality and safe products to the
consumers.
A basic machine vision system used in agricultural applications consists of a
camera, a computer with an image acquisition board, a lighting system and computer
software for capturing, storing, analyzing and processing the image.
When a light source illuminates the object, the incident radiation is transmitted,
reflected or absorbed by an object based on the optical properties of the object. These
optical properties are functions of wavelength, angle of incident light and the physical
and chemical composition of the object. The ranges of light used include Ultra Violet
(200-400 nm), Visible (400-700 nm) and Near Infra Red (700-2500 nm). The frame
grabber having features for image acquisition, control for camera and image data
preprocessing is important for any machine vision system. Most of the machine vision
systems use monochrome frame-grabber boards with RS-170 or CCIR video input or

colour frame-grabber which receives NTSC, PAL or S-VHS inputs.
3.3.2. Applications in agri-food industry
Machine vision has great potentials in agricultural sector because of its simplicity,
low cost, rapid inspection rate and non-destructive nature. Brosnan and Sun (2004)

8


reviewed the computer vision techniques for improving the quality inspection of food
products. They discussed in detail about the basic concepts, advantages and demerits of
the machine vision systems. The research works carried out by different researchers and
recent developments in quality detection in bakery products, meat and fish, vegetables,
fruits, prepared consumer foods, grain, food container inspection and other applications
were also discussed.
Chen et al. (2002) reviewed the application of machine vision in agriculture and
detailed about the requirements for a machine vision system, latest developments in
hardware and software for the system with special emphasis to multi and hyperspectral
imaging for food inspection. In their case studies they enumerated the adoption of
hyperspectral imaging system having a CCD camera with a control unit interfaced with a
computer and light source with 150W halogen lamps. An experiment for detecting
contaminated Red Delicious apples was conducted. They concluded that a simple ratio of
images of fluorescence bands at 680 and 450 nm differentiated the uncontaminated
surfaces from contaminated portions irrespective of the skin colour.
A machine vision system for automatic inspection of corn kernels was designed
and constructed by Ni et al. (1997). The developed classification algorithm was tested for
on-line classification of whole and broken corn kernel. The system classified the whole
and broken corn kernels with an accuracy of 91% and 94%, respectively.
3.4. Visible / near infrared spectroscopy (VNIRS)
3.4.1. Back ground
Visible / NIR spectroscopy has been recognized as a rapid and non-contact method

for the determination of different constituents in agricultural and food products. Near-

9


infrared spectroscopy (NIRS) measures diffusely reflected or transmitted light over a
range of invisible wavelengths longer than the visible light within a small field of view of
the object. A spectrophotometer, a fiber optic reflectance probe, a light detector, a
lighting unit and a computer system with data acquisition software form a basic
spectroscopy system. In the system, the incident light may travel into the sample and
reflected. Visible/NIR detector records the intensity of reflected light at certain
wavebands from any particular point on the sample. The variations of the intensity at the
wavebands form a spectrum which could be used for further analysis.
3.4.2. Applications in agri-food industry
Near infrared reflectance spectroscopy (NIRS) in the wavelength range of 11002500 nm was used to predict the nutritive quality of corn silage by Lovett et al. (2005).
Adamopoulos and Goula (2004) adopted NIR spectroscopy to predict moisture, protein
and fat content of taramosalata a famous Greek dish and obtained the calibration results
with low standard error of prediction compared to the chemical methods for all the
components. Shackelford et al. (2004) conducted experiments to develop an optimal
protocol for using visible and near-infrared reflectance spectroscopic technique in the
wavelength range of 350 - 2500 nm to evaluate the longissimus quality traits of beef
carcasses. Wu and Shi (2004) studied the possibility of using NIRS for simultaneous
analysis of grain weight, brown rice weight and milled rice amylose content in rice
grains. A wavelength range of 1000 – 2500 nm was used.
Moisture content, oil and crude protein content on intact sunflower seeds were
examined using NIRS by Fassio and Cozzolino (2004). The samples were scanned in the
spectral range of 400 – 2500 nm and calibration equations were developed using

10



modified partial least square regression (MPLS) with internal cross validation. Harbeck et
al. (2004) applied NIRS with a spectral range of 1250 - 2200 nm for betaine detection in
sugar industry during desugarization process from molasses. Hans (2003) determined
two broad peaks responsible for water near 1440 and 1930 nm in any NIR spectrum and
described the influence of hydrogen bonding and temperature of the sample.
Xing et al. (2003) adopted visual (VIS) and near-infrared (NIR) spectroscopy to
discriminate bruised and non-bruised healthy spots on ‘Golden Delicious’ apples. They
used this technique for detecting two types of bruises i.e. those created by controlled
impact and those by compression. Spectral reflectance was measured in the wavelength
range from 400 to 1700 nm.
Park et al. (1996) tested an integrated system which consisted of a visible/nearinfrared (NIR) spectroscopic subsystem and an intensified multispectral imaging
subsystem for its accuracy in separating abnormal (unwholesome) from normal poultry
carcasses. The poultry carcasses were measured at wavelengths from 471 to 965 nm by
spectroscopic subsystem and the multispectral imaging subsystem measured the graylevel intensity of whole carcasses using six different optical filters of 542, 571, 641, 700,
720 and 847 nm wavelengths. From the results they concluded that the integrated system
judged the abnormal carcasses to 100% accuracy where as spectroscopic subsystem had
an error of 2.6% and the multispectral imaging subsystem had an error of 3.9% being
misclassified as normal carcasses.
Story and Raghavan (1973) used the infrared reflectance technique for separating
stones and soil clods from potato in a potato harvester. They exploited the reflectance

11


bands of potato, stones and soil clods. Potato had reflectance ratio of 0.6 to 1.3 micron
where as stones and soil clods had 1.5 and 2.4 micron bands respectively.
3.5. Hyperspectral imaging
3.5.1. Back ground
Hyperspectral imaging is a new imaging technique that can be used to measure

internal quality of food materials, fruits and vegetables. It measures diffusely reflected or
transmitted light from visible to over a range of invisible wavelengths from a larger area
on the object.
Tatzer et al. (2005) developed an industrial online material sorting system with
spectral imaging technique. Functional components and classification methods was
studied for cellulose-based materials such as pulp, paper and cardboard. Patrick et al.
(2004) developed a hyperspectral imaging technique with a high spatial resolution
(0.5-1.0 mm). The system detected defective and contaminated foods and agricultural
products. Various image analysis methods were compared for the detection of defects
and/or contaminations like side rots, bruises, flyspecks, scabs and molds, fungal diseases
(such as black pox), soil contaminations on the surfaces of Red Delicious, Golden
Delicious, Gala, and Fuji apples. Differences in spectral responses were analyzed using
monochromatic images and second difference analysis methods for sorting wholesome
and contaminated apples.
3.5.2. Theory of hyperspectral imaging technique
Hyperspectral imaging technique combines spectroscopic and imaging systems to
collect spectral and spatial information simultaneously. The data collected can be

12