nondestructive estimation of relative contents of … · cell vollume increases too. si ilsimilar...
TRANSCRIPT
Shiraz University
Shiraz, Iran
Department Of Mechanics of Agricultural Machiner EngineeringDepartment Of Mechanics of Agricultural Machinery Engineering
NONDESTRUCTIVE ESTIMATION OF RELATIVE CONTENTS OF CHICKEN EGG BY MACHINE VISION AND
NEURAL NETWORK TECHNIQUESNEURAL NETWORK TECHNIQUES
M. H. RAOUFAT
V. ASADI
Egg-producing countries worldwide show a growing trend.Egg world production in 2005 was more than 64,394,000tons and in 2006 showed an increase 1/7 percentage with65,500,000 tones.Iran having the fifteenth rank global and fifth in Asian,b f l b l d i fabout one percent of global production of eggs.
tion
oduc
tIntro
Egg is a rich source of protein, nutrients (sodium,potassium, calcium, phosphorus, magnesium, iron, zinc,po ass u , ca c u , p osp o us, ag es u , o , c,copper, iodine, chlorine, sulphur and selenium) andvitamins (A, B1, B2, B3, B6, B12, D and E). Egg( , , , , , , ) ggcontains 65-70% moisture, 11-12.5% protein and 9.5-10.8% fat. Investigations have show that the protein ing pthe egg has a high nutrient value. This worthy nutritiveegg is capable of developing embryo and chicken baby.gg p p g y y
3
Nutrient content of an eggtion
oduc
tIntro
tion
oduc
tIntro
Egg white and egg yolk are extensively utilised asingredients because of their unique functionalproperties, such as gelling and foaming. Foams areused in the food industry for the production ofbread, cakes, crackers, ice creams, etc. Hen egg yolkhas good emulsifying properties.
6
Objectives:
To develop and evaluate a number of algorithms for To develop and evaluate a number of algorithms for estimating egg volume
To develop an algorithm for estimating relative contents of eggcontents of egg
To investigate the effects of storage time temperature To investigate the effects of storage time, temperature and moisture on relative contents of egg
esectiv
eObje
Literature review
Application of machine vision in agricultural engineering:
Quality control and grading of various products(Kondo et al. 2000 & Blasco et al. 2003)
Automatic guidance of agricultural machinery
tion
(Fehr & Gerish 1995. Brandon 1992, Gerish et al. 1997)
A i h i f d
oduc
t Automatic harvesting of products(Fitzpatrick et al. 1997 & Benson et al. 2003)
Intro
Forbes, K. (2000)
Forbes (2000) used radial photo images to Forbes (2000) used radial photo images to estimate surface and volume of apple, strawberry and tomato and quanta lope The pictures were and tomato and quanta lope. The pictures were captured using mirrors installed at 45o around fruit Then pictures were used to establish a three fruit. Then pictures were used to establish a three dimensional representation of the fruit, the data were used to predict surface and volume of the were used to predict surface and volume of the fruit using neural network technique. Although the method was accurate but it was time ti
on
the method was accurate but it was time consuming and therefore not suitable for real‐time applicationsoduc
t
time applications.
Intro
Lee, D. J., Lane, R. M. and Chang, G. H. (2004)
L t l ( ) d l b t d t t Lee et al (2004) used laser beam to detect any surface unevenness in the muscles and to establish th di i l i f th l Th a three dimensional image of the muscles. The image could be used to estimate the volume.
tion
oduc
tIntro
Diaz, R., Gil, L., Serrano, C., Blasco, M., Molto, E. and Blasco, J. (2004).
Diaz et al (2004) used combination of image processing and neural network techniques for processing and neural network techniques for grading of table olive. The image processing could identify the bruised areas on the fruit surface and identify the bruised areas on the fruit surface and give information on bruised area and color intensity. The data were then fed to a neural network. The The data were then fed to a neural network. The network could separate the fruits into four distinct groups different in quality.ti
on
groups different in quality.
oduc
tIntro
Blasco, J. Aleixos, N. and Moltó, E. (2007).
Blasco et al (2007) developed a region oriented ( 7) p gsegmentation algorithm to detect orange fruit surface bruise and also to differentiate fruit bruise from calyx. The algorithm was based on recognition of localized color intensities and was not merely ybased on single pixel color.
tion
oduc
tIntro
Studies on quality grading of egg can be classified into two groups:1. Those based on apparent defects and2. Those based on internal defects
The studies have been mainly conducted by adopting image processing techniques and sound ti
on
wave propagation.
oduc
tIntro
Das, K. and M. D. Evans. (1992).
Das and Evans (1992) used the image processing techniques to separate fertile from unfertile eggs. They captured images f d i h i b f d d Hi of eggs stored in the incubator for 3 and 4 days. Histogram
enhancement could segregate the samples into two groups; fertile and unfertile The accuracy for 3 and 4 days stored fertile and unfertile. The accuracy for 3 and 4 days stored samples were 89 and 98 %, respectively.They also trained a neural network system to analyze the y y yhistograms. They divided the histograms into 8 regions and considered the pixel intensities as network inputs. They
d i f d % f h d tion
reported a sorting accuracy of 93.5 and 93.9% for the 3 and 4 days stored eggs, respectively.
oduc
tIntro
Patel, V. C., Mcclendon, R. W. and Gooddru, J. W. (1994), (1998)
Patel et al. developed a color computer visionbi d ith l t k t fcombined with a neural network system for
detection of eggs with defect. They joined threehi t f h d f d hi thistograms of each egg and formed a histogramwith 768 cells. Each cell was considered as an inputt th l t k Th t bl fto the neural network. The system was capable fordirt stain detection with 97.8% accuracy, 91.1%
f bl d t d t ti d 96 7%tion
accuracy for blood spot detection and 96.7%accuracy for crack detection.
oduc
tIntro
Jindal, V. K., Sritham, E., (2003).
Cho et al (2000) used the sound wave propagation h l d h h ll k htechnology to detect the egg shell cracks. The system
accuracy was 95% and could detect cracks at the rate f f hof 200 ms for each egg.
Jindal et al (2003) continued the study and could d h h ll k h hdetect the shell cracks at higher accuracy.
tion
oduc
tIntro
Nakano, K., Motonaga, Y. (2003).
Nakona and Motonaga (2003) used the ultra i l d NIRS d h i h bl d violet and NIRS to detect the eggs with blood spots . The blood could interact with the wave hi h ld b d i f bl d Th which could be used as sign of blood spots. The
system could detect the healthy eggs with 100% d h i h bl d i h 6 8% and the eggs with blood spots with 96.8%
accuracy.
tion
oduc
tIntro
Evaluation of egg freshness
Estimating egg quality (Freshness):(Freshness):
Destructive techniques
tion
Destructive techniques
Non‐destructive techniques
oduc
t q
Intro
Destructive techniques for egg freshness evaluation
Destructive methods:Haugh UnitThe most widely method for estimating egg quality is the Haugh Unit whichis calculated based on the egg weight and the albumen height of a brokengg g gegg.Air cell heightAir cell height is the only quantitative egg freshness parameter consideredAir cell height, is the only quantitative egg freshness parameter consideredby the European Union regulation. Theoretically, a grade A egg at packaginghas to keep the characteristics of its grade (air cell height <6 mm) up toexpiring dateexpiring datepH of albumenThe pH of albumen from a newly laid egg is between 7.6 and 8.5. Duringh f h ll h H f lb iti
on
the storage of shell eggs, the pH of albumen increases at a temperature-dependent rate to a maximum value of about 9.7.
oduc
tIntro
Nondestructive techniques for egg freshness evaluation
Non‐destructive methods:
Berardinelli et al (2005) used the infrared waves to evaluate the albumin height, this method could estimate the height up to 1mm±1mm
Abunajmi et al (2010) used the ultraviolet waves to estimate the freshness (age) of eggs. They stored 300 estimate the freshness (age) of eggs. They stored 300 eggs at 4 and 300 eggs at 25 oC for 6 weeks. The study revealed that as the egg freshness decreases, the speed ti
on
of wave propagation through egg decreases too.`
oduc
tIntro
21
Development of Image Acquisition system (Non-destructive approach).
Main parts:light source main compartment egg holder and cameraod
s
light source, main compartment, egg holder and camera.For non-destructive observation of internal parts of egg a highintensity light beam is necessary to penetrate into egg membrane m
etho
y g y p ggthrough its longer axis. In this system a halogen bulb with yellowlight was used. light source was placed inside the chamber to
f li h i h di i h ld i h and m
prevent scatter of light in other directions. Egg holder in themain compartment was a horizontal plane with a hole 5 mm india through which light beam pass and penetrate into eggrials a
dia. through which light beam pass and penetrate into egg.
Mate
ods
metho
and m
rials a
Mate
Egg storage conditions
Temperature levels5 oc202030
ods
metho
and m
rials a
Mate
Relative humidity levels, %
ods
metho
and m
rials a
Mate
Estimating egg volumeod
s m
etho
and m
rials a
Mate
Egg volume calculation methods
1. Ellipsoid approximation
ods
S f di k l t metho
2. Sum of disk elements
and m
rials a
Spiegel, M., Lipschutz, S. and Liu, J. (2008)
Mate
f l ( )3. Sum of cone elements (SCE)
Flowchart for egg content volume estimationod
s m
etho
and m
rials a
Mate
Flowchart for estimating egg contents volumevolume
ods
metho
and m
rials a
Mate
Artificial Neural Networkod
s m
etho
and m
rials a
Mate
Artificial Neural Networkod
s m
etho
and m
rials a
Mate
Laboratory examinations
Egg volume measurement. Three mentioned methods were used to estimate egg volume. Control measurement was conducted in the lab.s
thod
s
Egg components volume measurement. Sum of disk element method was used to estimate d
me
component volumes using two images for each data point.
als a
ndateria
Ma
RESULTS & DISCUSSIONRESULTS & DISCUSSION
34
Evaluation of egg volume algorithms
Diff b t i th d Difference between various methods
ion
scus
sind dis
lts an
Resu
l
Effect of number of views on estimated volume
ion
scus
sind dis
lts an
Resu
l
Difference between calculated volume and estimated volume
EV: ellipsoid method,SDE: sum of disk elements,io
n
SCE: sum of cone elements,
MV: measured volume
scus
sind dis
lts an
Resu
l
Performance of algorithmsion
scus
sind dis
lts an
Resu
l
Bland–Altman plot for the comparison of measured and ti t d l t li i di t th 95% li it festimated volumes outer lines indicate the 95% limits of
agreement.SDE2SDE2
sion
scus
snd di
ults an
Resu
Evaluation of the algorithm for estimating contents Evaluation of the algorithm for estimating contents volume
ion
scus
sind dis
lts an
Resu
l
ion
scus
si
Air cell Albumen
nd dis
lts an
yolk
Resu
l y
Investigate the effect of time, moisture and content on content egg
Change of air cell volume during storage time
ion
scus
sind dis Moisture 11% Moisture 32%
lts an
Resu
l
Moisture 52%
Change of albumen volume during storage time
ion
scus
si
Moist re 11%
nd dis Moisture 11% Moisture 32%
lts an
Moisture 52%
Resu
l Moisture 52%
Change of yolk volume during storage time
ion
scus
si
Moist re 11%
nd dis Moisture 11% Moisture 32%
lts an
Moisture 52%
Resu
l Moisture 52%
Conclusions
11. No significant difference was noticed among h l i i h d l h h hthree volume estimating methods, although the SDE could predict egg volume more accurately.
12. Use of two views for image analyses1Of h l i ld t 1Of each egg sample yields more accurate
results although not necessarily in a significant manner
45
Ad i l ifi i d l 3‐ Adopting color specifications and Neural network technique can help separate three regions i id ( lk lb d i ll) Th inside an egg (yolk, albumen and air cell). The volumes can be estimated using SDE method.
4. As egg life and air temperature increases, the air ll l i Si il i i i cell volume increases too. Similar increase in air
cell volume is noticed for decrease in humidity. Y lk l d lb d h i Yolk volume and albumen decreases as the air temperature increases. These findings are i di i f f h
46
indications of egg freshness.
47