Terrain
classification for track-driven agricultural robots
Khairul Azmi Mahadhira, Shing Chiang Tanb, Cheng
Yee Lowa,c
, Roman Dumitrescud, Adam Tan Mohd Amina, Ahmed
Jaffara,c
Abstract
Agricultural automation uses intelligent robots to
assist with challenging tasks such as managing crops or harvesting less effort.
This requires agricultural robots to be able to work in different field
conditions. The reason is that field features can affect the performance of the
robot while working on the field, capturing it on the ground. The purpose of
this work is to evaluate efficiency using intelligent algorithm, ie vector
support (SVM), when different field conditions are introduced into the
agricultural area. For this purpose a prototype of a portable slider robot was
developed, along with various terrains. The head of the test shows three
different types of field conditions, namely sand, gravel and vegetation. The
track robot is also connected to a measured MEMS meter to measure vibrational
signals generated by field simulations. The experiment was carried out using
three-function trained SVM, ie, Linear Function, Polynomial Function and Basic
Radial Function (RBF). The results showed that SVM could very well recognize
the different terrain conditions. This work contributes to adapting to a farmer
with changing the terrain.
1.
Introduction
Many jobs are needed in the traditional agriculture
industry. In recent years, a large number of robots have been developed to
assist human activists in agricultural activities. For example, robots are used
for fumigation, fruit harvesting and transport modification 1-3. Therefore,
recent research on software has enabled robots to study environmental data and
have the ability to adapt to the environment 4-5. One approach is the
placement of machine learning techniques 6.
In the field of agricultural automation, robots can
be equipped with computer vision systems that perform visual navigation
For example, low-cost robots have vision control
systems that provide visual navigation for artificial pollination and
fumigation in a greenhouse environment 8. Artificial vision systems are also
installed in robots to move between rows of crops 10 automatically recognize
the status of the fruits before harvest or 11 sort fruit. On the other hand,
there are studies on the fusion of automatic learning technology and
agricultural robots 12-13. For example, a collection robot 14 is installed
along with an automatic statistical learning method to recognize the maturity of
apples. The computer vision system integrates with artificial neural networks
to perform license image classification for sunflower crops. In this image,
applications can facilitate control of weeds 15-16.
In agriculture, field conditions
through which these robots affect the performance of robots in their work. For
example, weighing robots produce high vibrations for robots on these surfaces.
In this study, SVM (Vector Support Machine) technology was proposed as a
learning algorithm to differentiate the various field conditions in
agriculture. To evaluate the effectiveness of the algorithm, the atrack powered
mobile robot has a MEMS accelerometer built to measure vibration data, which is
analyzed and categorized using SVM. Knowing ground conditions, machine controls
can be applied to generate the moment the robot needs to move when the field
conditions change in the field.
The document is structured in
this way: Section 2 describes the architecture of a robust farm-based system.
Section 3 describes the theoretical background of the supporting vector machine
used for field classification. Section 4 presents the results of field field
experiments. The conclusion is made in 5.
2. System architecture of a track-driven agricultural robot
The purpose of this study was to evaluate the
recycling of the SVM and the recognition of land on agriculture, ie sand ,
gravel and soil. For this purpose, a small robot is developed with a field test
site. A key element of the agricultural robot system developed for this work is
indicated in the active structure 17 at the top of figure 1. The behavioral
diagram is assumed to be in the bottom of Figure 1.
As in the active structure in 1, the rocket robot
of a DC motor is indicated and Robot maneuver with another drive system 18. A
low and low MEMS accelerometer (ADXL335 from Spark Fun Electronics) is
integrated into the robot to report vibrations caused by fishy interactions.
Based on vibration gain SVM signals and defines a kind of field, ie sand, soil
or gravel. According to the ground, the fineness (FSM) 19 is used to adjust
the behavior of the system by causing a specific state of the system system .
The controller then signals control signals for electric power to change the
input and motor voltages.
As shown in the status scheme of Figure 1 when the
state of the system is triggered for flooding, the DC motor is controlled to
generate low revolutions and torque in a steady robot. When capturing the
system, DC motors are used to produce high speeds at low speeds. When the
system comes out of the sand, the DC motor is received to produce more
pressures at lower speeds. Since field conditions and system behavior are
recognized, agricultural robots repair terrain without being caught in the
field.
Fig.1. Active structure (upper part) and
behavior–state diagram (lower part) for a track-driven agricultural robot.
Figure 2
presents a data test that is controlled by three different types of terrain
over a continuous period of time. Before the classification declaration is
made, the accelerometer data has been filmed for vengeance. The resolution used
in data collection is about 100 Hz. Software components are developed in the
programming languages MATLAB and C.
Khairul Azmi Mahadhir et al. / Procedia
Technology 15 (2014) 776 – 783
779
Sand
Gravel
Soil
Robot
Fig. 2.Example of vibration
signal measured on different types of terrain, i.e. sand, gravel and soil.
3.
Support Vector Machine for
terrain classification
SVM is a computer method based on statistical
learning and optimization theory 20. It was chosen to classify the soil based
on information expression in sustainability and boundary classes 21. In the
SVM training process, he has a variety of hyperplaners that maximize the
difference from the next sample of any class size so that hyperplays can be
partially broken down through data classification. An example of a linear SVM
is shown in Figure 3. The SVM is set up for two classes of data that were
originally isolated
wtx
+ b = 0 (1)
watx is a prerequisite for data samples, vector
weights, and continuous offsets.
.
780 Khairul Azmi Mahadhir et al. / Procedia
Technology 15 (2014) 776 – 783
wt x + b = 1
wt x + b = -1
wt x + b = 0
Fig. 3.
Separating hyperplane in the SVM between two data sets.
In many cases, actual data is
complex. Linear SVM systems can not be effective in the efficient isolation of
these complex data that are not linear. An approach is a southern approach to
dealing with unsafe problems. Another way to overcome this limit on the SVM
model is to create unparalleled workforce to improve nonlinear data space transformations
to improve recognition. In this case, Keel Tricks 22, such as radial core
functions, polynomial errors, etc., can be determined by linear and nonlinear
classification.
SVM has two strategies for
classifying data problems and multiple classes (eg one-against-one (OVO) or
one-versus-all (OVA)). The OVO strategy was introduced for the first time in
the SVM 23 and is also known as a party or round building. This is actually
the basic form of binary classification. Let’s call the data sides.
D
{xm , ym }, m 1,…, n are available for training, where xm
ƒp is a
feature vector Showing the m
sample,
and ym {1,2,…K} is the class label of xm . The SVM model that
implements OVO will consist of K (K 1) / 2 binary SVMs. On the other hand, the OVA strategy is applied to
build K SVMs where the i-th SVM
is
trained with all the data samples of the i-class coded as 1, and the data
samples of other classes coded as -1. In this work, the SVM model is built to
solve a problem by using an OVA strategy, as follows.
whereC is a already defined parameter being
introduced according to a soft margin approach and it controls the trade-off
between training accuracy and generalization (note: an example of the effect of
C on a linear SVM is illustrated in
Figure 4; w j is the
weight vectors of SVM trained with data samples from two classes; I( x
j ) is the kernel
function; bi is a scalar; ij
is the
slack variable that permits i 1, , ns constraints to be
violated; z j {1,1} is the class label for the classifier. Given a data sample x, the decision function
of the SVM is
(a) C=10
(b) C=100
Figure 4. Effects of a slight
constant difference C. The figure shows that C = 10 left (a) and C = 100 in the
right image (b) Figure shows that positive and negative samples can be divided
into hyperplanes. (b), the hyperplane is closer to the boundary, while the
difference increases. By selecting the appropriate value for parameter C, the
SMV can be performed with optimal classification results by reducing the number
of training errors. 23
4.
Classification results
In the experimental configuration
are the three areas of agriculture concerns d. Sand, gravel and soil. The data
were measured by the accelerometer with three axes, and 10 ZAXIS xaxis compared
to 2G. One experiment used war, our data (such as gravel) Zaxisis vertical
acceleration is used “single” and additional information (sand
floor), are used together, and that’s all. Then, by means of a test of the sand
of the ‘one’ and the second combining ‘every’, is, and for the ground which the
rocks. To the car is said to do, came out against the maintenance of a vector
in the hyperplane at the distance 10 times during the experiment, ie a part of
a given 10 parts of bare gravel is placed as the next training tests. That was
the experience of a given than the ground, and out of the wall of sand. In the
process of experiments after three thousand measures. As shown in Table 4, the
numbers are linear function, muscle function, and the entire basic function
(RBF).
Table 1. Percentage of
Classification Rate and Mean for X-Y-Z-axis
using different kernel functions.
Type of terrain
Gravel
Sand
Soil
Axis
X
Y
Z
X
Y
Z
X
Y
Z
Kernel Function
Linear
Test
40%
34%
70%
89.6%
88%
53.5%
48%
60%
64%
Mean
14.14
18.37
13.72
7.76
2.88
11.47
19.23
10
20.02
RBF
Test
80%
77%
93%
92.4%
83%
92.4%
79.8%
80%
84%
Mean
10
9.48
11.41
4.85
5.77
11.81
15.49
7.42
15.85
Polynomial
Test
44%
32%
46%
94.8%
86.7%
86%
44%
60%
53%
Mean
8.94
20.44
15.06
5.59
5.77
11.43
20.74
10
20.02
For the first approach, the
experiment uses only one vertical acceleration data along the Z-accelerated
acceleration axis. The experiment shows that the degree of cleavage at Zaxis
has
the highest degree of
classification compared to other terrain, and is expected in the results of the
high acceleration acceleration obtained from accelerometers to gravel. The
experiment shows that by altering the kernel function the learning algorithm
creates another result. For learning processes, a data set and the remainder for
the training are used for testing, and the result is visible in Table 1. The
RBF results show higher stability in the evaluation compared to the linear
functions of the kernelsin polynomial functions on the Z axis.
The second approach also used the
ONE versus ALL strategy for linear function, radial base function (RBF) and
polynomial function compared to Yaxis run for the whole terrain and then only
for the X axis. Data obtained from the Support Vector Machine using Yaxis and
Xaxisis do not correspond to the Z axis in the data collected by horizontal and
lateral acceleration do not represent real vibrations from the surface of the
terrain.
The mean values ??obtained in
Table 4 show the data variation from the mean values. The lower deviation
determines numbers that approximate the average or expected data, and the
higher the deviation, the numbers are spread over a wide range of values.
5. Conclusion
Vector support vectors are built
on the boards of agriculture agricultural tractors to determine the terrain
function classification. Vibration signals due to field terrain interaction are
measured and used as inputs to vector vectors. The efficiency of the three
kernel functions, the linear function, the polynomial function, and the radial
basis function (RBF) were compared using 10-fold cross-certification. The
results show that the SVM training model of RBF can better identify terrain
conditions than linear functions and polynomial nuclei.
Acknowledgements
The authors thank the Ministry of Science, Technology and Innovation Malaysia and the Ministry of Education Malaysia for funding the research work
through ScienceFund Ref. 100-RMI/SF
16/6/2 (3/2014) and Exploratory Research
Grant Scheme Ref. 600-RMI/ ERGS 5/3 (13/2013). The authors thank the
Research Management Institute of
Universiti Teknologi MARA for managing the research funds.
References
1 Allotta,
B., G. Buttazzo, P. Dario and F.A. Quaglia. Force/torque sensor-based technique
for robot harvesting of fruits and vegetables. Proceedings. IEEE International
Workshop on Intelligent Robots and Systems (IROS) ; 1990 , p. 231-235.
2
Edan, Y.,
D. Rogozin, T. Flash and Miles G.E. Robotic Melon Harvesting. IEEE Transactions
on Robotics and Automation 2000:831–835.
3
Sistler,
F. Robotics and intelligent machines in agriculture. IEEE Journal of Robotics
and Automation 1987:3–6.
4 Gausemeier
J, Kahl S, Low CY, Schulz B. Systematic Development of Controllers based on the
Principle Solution of Self-Optimizing Systems. In Proceedings of the DESIGN
2008, 10th International Design Conference, Dubrovnik, Croatia, 2008;
1263–1276.
5 Low CY,
Aldemir M, Aziz N, Dumitrescu R , Anacker H , Mellado M. Strategy Planning of
Collaborative Humanoid Soccer Robots based on Principle Solution. Prod. Eng.
Res. Devel. 2013; 7:23–34.
6 Mousazadeh
H. A Technical Review on Navigation Systems of AgriculturalAutonomous Off-road
Vehicles, Journal of Terramechanics 2013:211–232.
7 Santosh
A, Gerie W.A.M. van der Heijden, Frits K. van Evert, Alfred S, Cajo J.F. ter
Braak. Laser range finder model for autonomous navigation of a robot in a maize
field using a particle filter. Computers and Electronics in Agriculture;100:
41-50.
8 Belforte
G, Deboli R, Gay P, Giglio A.Robot Design for Applications in Intensive
Agriculture. 2002 IEEE International Conference on Industrial Technology; 2002,
p. 519-523.
9 Zhao
C.-J., Jiang G.-Q, Baseline detection and matching to vision-based navigation
of agricultural robot, in International Conference on Wavelet Analysis and
Pattern Recognition; 2010,p. 44-48.
10 JiW, Zhao
D, Cheng F, Automatic recognition vision system guided for apple harvesting
robot. Computers and Electrical Engineering 2012:1186-1195.
11
Roseleena J, Nursuriati J, Ahmed J, Low CY.
Assessment of Palm Oil Fresh Fruit Bunches using Photogrammetric Grading
System. International Food Research Journal 2011; 18(3).
12
Mousazadeh
H. A technical review on navigation systems of agricultural, Journal of
Terramechanics 2013:211–232.
13
Huang Y, Lan Y, Thomson SJ, Fangc A, Hoffmann WC,
Lacey RE. Development of soft computing and applications in agricultural and
biological engineering. Computers and Electronics in Agriculture 2010:107-127.
14
Wang J, Zhao D-A, Ji W, Tu J-J, Zhang Y.
Application of support vector machine to apple recognition using in apple
harvesting robot, IEEE Int. Conference on Information and Automation; 2009, p.
1110-1115
15
Young
J.R. Neural Network Control for Visual Guidance System of Mobile Robot. Lecture
Notes in Computer Science 2007: 685-693.
16
Astrand B, Baerveldt A. J. An Agricultural Mobile
Robot with Vision-Based Perception for Mechanical Weed Control. Journal of
Autonomous Robots 2002:21-35.
17
Gausemeier J, Low CY, Steffen D, Deyter S.
Specifying the Principle Solution in Mechatronic Development Enterprises. In:
2nd Annual IEEE Systems Conference, Montreal, Canada; 2008, p. 1–7.
18
Low CY,
Rosdayanti F, Azmi K, Che Zakaria NA. Steering Behavior of a Track-Driven
Paintball Robot. Procedia Engineering 2012.
19
Kohavi Z.
Switching and Finite Automata Theory. 2nd ED. New York: McGraw-Hill; 1978.
20
Vapnik V.
The Nature of Statistical Learning Theory. London: Springer-Verlag; 1995.
21
Weiss C, Holger F, Andreas Z. Vibration-based
Terrain Classification Using Support Vector Machines.Proceedings of the 2006
IEEE/RSJ International Conference on Intelligent Robots and Systems;2006,
p.4429-4434.
22
Shawe-Taylor
J,Cristianini N. Kernel Methods for Pattern Analysis. Cambridge: Cambridge
University Press; 2004.
23
Hsu CW, Lin CJ. A Comparison of Methods for
Multiclass Support Vector Machines. IEEE Transactions on Neural Networks
2002:415-425.
