Академический Документы
Профессиональный Документы
Культура Документы
b Pontifical
ABSTRACT
This paper reports the process of pattern classification of the chewing process of ruminants. We propose a
simplified signal processing scheme for optical fiber Bragg grating (FBG) sensors based on machine learning
techniques. The FBG sensors measure the biomechanical forces during jaw movements and an artificial neural
network is responsible for the classification of the associated chewing pattern. In this study, three patterns
associated to dietary supplement, hay and ryegrass were considered. Aditionally, two other important events
for ingestive behavior studies were monitored, rumination and idle period. Experimental results show that the
proposed approach for pattern classification has been capable of differentiating the materials involved in the
chewing process with a small classification error.
Keywords: Fiber Bragg gratings, Pattern Classification, Biomechanics, Artificial Neural Network
1. INTRODUCTION
The evaluation of the forage intake of ruminants is essential for understanding the grazing system dynamics.1
Procedures for estimating consumption include direct and indirect methods. In direct methods, forage intake
is evaluated based on digestibility measurements such as: ratio techniques that involve the calculation of digestibility and fecal output data; and index procedures that generally relate the level of intake or digestibility to
some component in the fecal matter.2 Indirect methods are based on direct behavioral observation or analysis
of acoustic signals and videos produced during the chewing process.3, 4 Acoustic analysis is the main method
used for monitoring the ingestive behavior of ruminants.5 However, the data collected with this method can be
corrupted by audio noise and the data classification is frequently performed without any automated technique,
which can be a difficult and time-consuming process.
In order to improve the analysis of chewing behaviour in ruminants, FBG sensors can be applied for data
acquisition and machine learning techniques for data classification. The use of FBG sensors to monitor the
ingestive behavior of ruminants allows significant improvements in the acquisition and processing of data.6, 7 In
addition, optical sensors are totally biocompatible;8 the sensor material (silica) is not toxic and is not rejected
by the animal body; it does not suffer from electromagnetic interference; it has diameter of micrometer order
and millimeters in length; and it provides excellent sensitivity in small signal acquisition. These characteristics
also allow the use of FBG sensors for monitoring irregular regions, e.g., facial bones.8
Machine learning techniques allows to build systems capable of acquiring knowledge automatically, using
existing datasets to improve their performance in a classification problem.9 The purpose of the present work is
to classify chewing patterns in ruminants using machine learning with in vivo data acquired from FBG sensors
installed in the jaw bone of a steer.
Further author information: (Send correspondence to Vinicius Pegorini)
Vinicius Pegorini: E-mail: vinicius@utfpr.edu.br, Telephone: +55 46 3220 2591
Rafael Cardoso: E-mail: rcardoso@utfpr.edu.br, Telephone: +55 46 3220 2575
24th International Conference on Optical Fibre Sensors, edited by Hypolito Jos Kalinowski,
Jos Lus Fabris, Wojtek J. Bock, Proc. of SPIE Vol. 9634, 963427 2015 SPIE
CCC code: 0277-786X/15/$18 doi: 10.1117/12.2195642
Proc. of SPIE Vol. 9634 963427-1
average value of one period of the signal was padded in order to complete 1000 attributes of one instance. The
16 remaining attributes were formed by the frequency components of a signal obtained from the FFT.11
For the training and validation phases of this ANN, based on the acquired mastication signals, it was defined
a dataset of input values for the ANN. The 1000 instances of the dataset was divided in two parts. The first 700
chewing movements were used in the ANN training phase, and the other 300, for the validation.
-E" 1541.25
1541.25
1541.25-
L 1541.20
r 1541.20-
15)
1541.20
1541.15
1541.15
ai
g 1541.10
0 1541.10
1541.05
0
1541.05
1541.05
1541.15
0 1541.10
2
Time (s)
(c) Ryegrass.
(b) Hay.
1541.25 -
Time (s)
Time (s)
1541.25
r 1541.20-
r 1541.20
15i)
ai
1541.15
ai
1541.15
g 1541.10 + 0 1541.10
7)
1541.05
1541.05-
2
Time (s)
Time (s)
(d) Rumination.
Figure 1. Chewing signals: (a) dietary supplement, (b) hay, (c) ryegrass, (d) rumination and (e) idle period.
200
0
20.0% 0.0%
0
199
3
2
0.2%
0
2
0.2%
9
93.9%
0.9% 16.1%
4
98.0%
0.4% 2.0%
0.0%
0.0%
0.0%
0
0.0%
0
0.0%
0
0.0%
21
0.1%
2.1%
0.0%
12
90.4%
1.2% 9.6%
0.6%
192
8
2
95.0%
0.8% 19.2% 0.2% 5.0%
173 88.7%
7.3% 11.3%
1 Dietary supplement
2 Hay
3 Ryegrass
4 Rumination
5 Idle period
Target Class
In the confusion matrix of Figure 2 , the diagonal cells, depicted in green, show the number of correctly
classified cases, and the cells above and below the diagonal, depicted in yellow, show the number of wrongly
classified cases. The blue cell at the lower right corner shows the percentage of the correctly classified cases
(presented in green), and the wrongly classified percentage is depicted in red.
The output of the ANN is acceptable and can be evidenced by the high number of correct responses and
also by the low number of wrong responses. For the case under study, with a number of five classified chewing
patterns, the neural network provided a percentage of 93.3% of correct classifications.
Dietary supplement, hay and rumination had the best classification results. The samples of these materials
had different wavelength values and also different waveforms. It provides the best classfication results. The
pelleted concentrate diet consists of small grains, which facilitates the food intake by the animal, requiring lower
jaw forces involved. Hay has a more rigid structure, because it was provided dehydrated, whereas during the
rumination the structure of the bolus is pasty. The ryegrass and idle period had the lowest rates. This is related
to the characteristics of each class. During idle period, the animal applied less mandibular force, because there
was no food between the dental arch. Ryegrass was given in its natural condition, so this material had a high
concentration of water, resulting in a softer material.
4. CONCLUSIONS
In this study, the induction rule obtained was able to generalize most patterns appropriately. The results showed
that the technique can be used to generate a trained neural network classifier with chewing data for different
types of food. The classifier can also be useful for identifying other animal nutrition-related events, such as
idle period and rumination, which are important data for studies related to animal health and welfare. The
FBG sensores provides high sensitivity while it is immune to electromagnetic interference, and has excellent
biocompatibility.
ACKNOWLEDGMENTS
The authors are grateful to CAPES, CNPq, FINEP, Araucaria Foundation and UTFPR by finantial support.
REFERENCES
[1] Ungar, E. D., Ingestive behavior, Int: Hodgson, J; J. Illius, A. W. (Ed.) The Ecology of management of
grazing systems. 1, 185218 (1996).
[2] Cordova, F. J., Wallace, J. D., and Pieper, R. D., Forage intake by grazing livestock: a review, Journal
of Range Management 31, 430438 (1978).
[3] Griffiths, W. M., Alchanatis, V., Nitzan, R., Ostrovsky, V., Ben-Moshe, E., Yonatan, R., Brener, S., Baram,
H., Genizi, A., and Ungar, E. D., A video and acoustic methodology to map bite placement at the patch
scale., Applied Animal Behaviour Science 98, 196215 (2006).
[4] Laca, E. A., Ungar, E. D., Seligman, N. G., Ramey, M. R., and Demment, M. W., An integrated methodology for studying short-term grazing behavior of cattle., Grass Forage Science 47, 8190 (1992).
[5] Laca, E. A., Ungar, E. D., and Demment, M. W., Mechanisms of handling time and intake rate of a large
mammalian grazer., Applied Animal Behaviour Science 39, 319 (1994).
[6] Wosniack, C., Silva, W. J., Cardoso, R., Assmann, T. S., Hill, J. A. G., Silveira, A. L. F., Sousa, K. M.,
Kalinowski, H. J., and Silva, J. C. C., Determination of chewing patterns in goats using fiber bragg
gratings, Proc. SPIE 8421, 84214F (2012).
[7] Karam, L. Z., Pegorini, V., Pitta, C. S. R., Assmann, T. S., Cardoso, R., Kalinowski, H. J., and da Silva,
J. C. C., Ex vivo determination of chewing patterns using fbg and artificial neural networks, Proc.
SPIE 9157, 91573Z191573Z4 (2014).
[8] Carvalho, L., Silva, J. C. C., Nogueira, R. N., Pinto, J. L., Kalinowski, H. J., and Simoes, J. A., Application
of bragg grating sensors in dental biomechanics, The Journal of Strain Analysis for Engineering Design 41,
411416 (jan 2006).
[9] Mitchell, T. M., [Machine Learning], McGraw-Hill, Inc., New York, NY, USA, 1 ed. (1997).
[10] Haykin, S., [Neural Networks: A Comprehensive Foundation ], International edition, Prentice Hall (1999).
[11] Cooley, J. W. and Tukey, J. W., An algorithm for the machine calculation of complex fourier series,
Mathematics of Computation 19(90), 297301 (1965).