IOP Conference Series: Materials Science and Engineering

PAPER • OPEN ACCESS

Identification of sugarcane maturity scale based on RGB, Gabor feature

extraction and Support Vector Machine
To cite this article: C Rahmad et al 2018 IOP Conf. Ser.: Mater. Sci. Eng. 434 012065

View the article online for updates and enhancements.

This content was downloaded from IP address 104.227.135.24 on 04/12/2018 at 17:21

3rd Annual Applied Science and Engineering Conference (AASEC 2018) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 434 (2018) 012065 doi:10.1088/1757-899X/434/1/012065
1234567890‘’“”

Identification of sugarcane maturity scale based on RGB,

Gabor feature extraction and Support Vector Machine

C Rahmad1, F Rahutomo1, M A Gustalika2* and I F Rahmah2

1
Information Technology Department, State Polytechnic of Malang, Malang,
Indonesia
2
Electrical Engineering Department, State Polytechnic of Malang, Malang, Indonesia

*[email protected]

Abstract. Sugarcane is a unique grass variety that allows many cuts for more than a few years.
Plant growth is very slow in the early stages that takes 25-30 days to complete germination and
90-95 more days to complete the tillers. Small scale sugarcane farmers generally do not have
direct access to sugar mills but through loggers. Sugarcane farmer must increase cost for the
loggers by themself. Sugarcane farmers often forgot the time at the moment when they planted
sugarcane. This study proposes a new approach to establish detection of maturity of sugarcane
before harvesting by detecting through stem of sugarcane. This study illustrates the maturity of
sugarcane. This study used its own datasets of 300 stem images of sugarcane and to differentiate
into three categories, immature, semi-mature, and mature. The study is divided into two
processes, the first one is extraction using the RGB color feature and Gabor texture feature, and
the second is the classification process using the Support Vector Machine, Naive Bayes and
Artificial Neural Network methods. Based on the trial it is known that Support Vector Machine
has the best results with precision and recall of 87,4% and 85,7% respectively.

1. Introduction
Sugarcane is a unique grass variety that allows many cuts for more than a few years and is considered
semi-lasting. It grows rapidly with abundant reproduction and allows for the financial management of
many parts of the plant. The expansion cycle consists of stages of sprouting, affiliation, enlargement and
maturation. From planting to primary cutting, the plant is called sugarcane, which lasts between 12 to
18 months, depending on the period and area of planting. After the first cut, the cane regrets were
replaced by cane rattan, with a standard 12-month cycle [1].
Sugarcane is a long-lasting plant that reaches its maturity in 11-12 months. Plant growth is very slow
in the early stages that takes 25-30 days to complete germination and 90-95 more days to complete the
tillers. Larger amounts of nutrients and water are applied during the early stages of growth compared to
the later stages. Sugarcane set is planted continuously in line with distance of 90 cm to 150 cm. Large
amounts of nutrients, moisture, solar radiation and wider distances between plants also support the
growth of weeds in sugarcane fields that prevent plants from getting nutrients [2].
Brazil is the world's largest sugarcane producer, followed by India and China, and is the largest
producer of sugar and ethanol. Brazil is responsible for about 20% of sugar production and 40% of the
world's exported sugar. Cane is also very important in terms of the environment as ethanol is one of the

Content from this work may be used under the terms of the Creative Commons Attribution 3.0 licence. Any further distribution
of this work must maintain attribution to the author(s) and the title of the work, journal citation and DOI.
Published under licence by IOP Publishing Ltd 1
3rd Annual Applied Science and Engineering Conference (AASEC 2018) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 434 (2018) 012065 doi:10.1088/1757-899X/434/1/012065
1234567890‘’“”

best alternatives to reduce emissions of gas pollutants, which is the fundamental cause of the greenhouse
effect [3].
In India, most farmers do not receive predictable crops for a variety of reasons. The cultivation of
agricultural products in principle believes in climatic conditions. Rain conditions also affect rice
farming. In this background, farmers inevitably require a wise guide to envisioning future efficiency
gathering plus investigations to be resolved in arrangements to help farmers harvest crops well in their
crops [1].
Knowledge of sugarcane varieties is a difficult thing to do. Data about this knowledge is complex
data. Currently, cane varieties can be done manually by sugarcane experts. This is because the
information about sugarcane is a qualitative descriptive semantic. This makes the annotation process
difficult for the general public as well as computer programs [4].
In practice small scale sugarcane farmers generally do not have direct access to sugar mills but
through loggers (contractors). Involvement of loggers in the sugarcane trade had an impact on the
increase in production costs borne by sugarcane farmers. Sugarcane farmers often forgot the time at the
moment when they planted sugarcane so they do not know the definite time in the harvest of sugarcane,
the sugarcane farmers only see from the condition of sugarcane only during the harvest. With this
problem, this study aims to make the application using machine learning algorithm is Artificial Neural
Network, Naive Bayes and SVM that was useful to know the maturity of sugarcane through stems with
certainty and also help increase income from sugar cane farmer.
In previous study, has been experiments of Precision sugarcane monitoring using SVM classifier [5],
which monitors parameters (temperature and humidity) and is responsible for the healthy growth of
sugarcane plants. The difference with the previous study is we make an application based on the maturity
of sugarcane through stem of sugarcane and using one type of sugarcane with code PS 86-2. then there
is a previous study on Feature Extraction for Identification of Sugarcane Rust Disease [6], which aims
to find the right features in identifying rust diseases in sugarcane by combining color and texture extract
features. Other studies have also mentioned that there is no system of monitoring for sugarcane products
widely. In a near real time yield estimation for sugarcane in Brazil combining remote sensing and official
statistical data [7], it was explained that they used remote sensing technology to estimate the yield of
sugarcane maturity. We also created our own datasets that can be used for further studies.

2. Material and method

2.1. System Description

Following system description in this study

Figure 1. System description.

2
3rd Annual Applied Science and Engineering Conference (AASEC 2018) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 434 (2018) 012065 doi:10.1088/1757-899X/434/1/012065
1234567890‘’“”

Figure 1 shows the description system we created. Our system description consists of two main phases:
the training phase and the classification phase. In the training phase, the cane dataset in segmentation
uses otsu then looks for its feature extraction, there are 2 feature extractions in use, color extraction and
texture. On color extraction using RGB while on textural extraction using Gabor. Next classification of
training data that has been in the extraction by using SVM. In the second phase is the classification by
using the dataset of the data testing, it will compare the results of the SVM classification of training data
so it will produce a percentage of maturity between the mature, half-mature and immature.

2.2. Machine based approach

Machine learning algorithms are the processes used to adapt the model to a dataset, through training or
learning. The model studied is then used against independent datasets. Basic machine learning
algorithms require a pair of documents - training and testing of data. The training phase includes self-
training using various images and then tested using test data. There are several Machine learning
algorithms, such as Artificial Neural Network (ANN), Naïve Bayes (NB) and Support Vector Machines
(SVM), data testing requires the dataset to produce classification.

2.2.1 Support vector machine. SVM is a machine-based learning approach based on statistical theory,
which can find the optimal solution of classification results with limited information about small sample
datasets. Unlike other machine learning approaches, SVM uses the idea of kernel functions to convert
nonlinear problems into linear problems, and reduce the complexity of mapping [8]. By using functional
transformation and kernel function optimization, the optimal problem can be converted into a convex
quadratic functional extremism problem. In theory, SVM is bound to get global optimal solutions SVM
Formula is:
1 𝛾
∅(𝝎, 𝜺) = 2||𝝎||2 + 𝑛 ∑𝑖=1 𝜀𝑖

s.t. yi g(𝑥𝑖 ) = yi(𝑤 𝑇 𝑥𝑖 + 𝑏)≥ 𝜺𝒊 − 1, 𝑖 = 1,2,3, … . . 𝑛 (1)

𝛾
where 𝜀𝑖 are slack variables and > 0 acts as a term of regularization.
𝑛

2.2.2 Naive Bayes. The Naïve Bayes classification is a probabilistic classification based on the Bayes
Theorem. For each sample, the possibility of the class is determined. It is assumed that all attributes are
independent [9]. NB is a popular machine learning tool for classification. Naive Bayes works very well
on various issues. Naive Bayes formula is :
Znb= argmax P(c) Π P (a|c) (2)
Explanation of symbols
a = word
c = category
P(a|c) = word probability in category c
P(c) = probability of category c
P(a) = probability of a word.

2.2.3. Artificial Neural Network. Artificial Neural Networks are trained in the image. The ANN
architecture is based on advanced feed networks. This architecture is trained in accordance with Back
propagation algorithm. The results obtained are further improved by using the Radial function network
Base The performance of the radial base network functions outperformed the ANN [10].

3
3rd Annual Applied Science and Engineering Conference (AASEC 2018) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 434 (2018) 012065 doi:10.1088/1757-899X/434/1/012065
1234567890‘’“”

Figure 2. Artificial neural network for classification of cane.

As shown in Figure 2. It consists of a set of interconnected neurons to map the input to output features.
An advanced three-nerve neural network is implemented with an input, hidden h and 1 neuron output
indicating the cane as mature or not. The learning of neural networks is monitored and the weights are
adjusted according to the back propagation procedure

2.3. Sugarcane sample

The sugarcane samples used in this study were (code name = 86-2) or person call it (62).A total of 300
samples, with maturity levels of immature, semi-mature, and mature, were picked in December 2017
in Wates, Kediri. For supporting sugarcane color and texture rating, then training dataset divede 70
dataset training into 3 category. The remnant is shared for data testing. Table 1 shows the details of the
training data and the data testing of the sugarcane dataset.
Table 1. The details of data training and data test of
Sugarcane.
Category Data Training Data Testing
Imaature 70 30
Semi Mature 70 30
Mature 70 30
Total 210 90
Table 1 shows the image in each category of size 156 x 208 pixels where the image has been through
the process of segmentation

2.4. Feature extraction

Extraction features are generally based on two pixel criteria, a). Equations and b) Proximity of pixel
values. The criteria for similarity are based on the gray range of features corresponding to the threshold
to separate background image data. So the feature of extraction can be interpreted as a process to
distinguish features that distinguish objects from other objects [11].

4
3rd Annual Applied Science and Engineering Conference (AASEC 2018) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 434 (2018) 012065 doi:10.1088/1757-899X/434/1/012065
1234567890‘’“”

2.4.1. Color feature. In order to detect whether the cane is immature, semi-mature, and mature, the color
statistical and color texture features for each Red (R), Green (G) and Blue (B) channels are extracted as
given below
1 𝑀
Color Mean(µ) = ∑ 𝑋𝑖
𝑀 𝑖=1
(3)

Sugarcane maturity is strongly associated with color. Therefore, the color feature is extracted from the
cane image to classify it as immature, semi-mature, or mature. The values of R, G, B of the extracted
and average images are calculated based on the following equation. Thresholds are selected by
experiment and R is average compared to threshold. If it is larger than the threshold, then the mature
cane is considered immature [12].
𝑅
Mean R = (4)
𝑁

𝐺
Mean G = 𝑁 (5)

𝐵
Mean B = 𝑁 (6)

Where, Mean R = the mean value of the red layer, Mean G = the mean value of the green layer, Mean
B = the mean value of the blue layer, R = Red pixel, G = Green pixel, and B = Blue pixels.

2.4.2. Texture feature. In this study using Gabor feature as texture extraction on sugarcane. This filter
has received considerable attention in processing the proposed image. Also, since these filters can have
optimal localization properties in both the spatial and frequency domains, they are perfectly suited for
texture segmentation issues. Gabor filters can be viewed as sinusoidal fields of specific frequency and
orientation, modulated by Gaussian envelope [13]. Gabor Filter formula is:
1 𝑥 2 +𝑦 2
ψ(𝝎,𝜃) (x,y)=2𝜋𝜎2 exp(− )exp(2ϳπ𝝎x) (7)
𝜎2

where j = √−1
Gabor filters focus on a certain frequency range. If the input image contains two different texture
areas, different local frequencies between regions will detect the texture in one or more image output
sub-filters. The basic Gabor function can perform space decomposition. Each Gabor Filter is determined
by basic Gabor [11].
2.5. Classification
Features extracted in the previous step are classified into 3 algorithms, naive bayes, artificial neural
networks, and support vector machines. Then from the three algorithms are compared, which has the
highest accuracy in determining the maturity of sugarcane. After that will get the results whether the
sugarcane is in the category immature, semi-mature, and mature.

3. Experiment and result

Detection of sugarcane maturity can be seen with the color difference in each class. Among the mature,
semi-mature, and immature classes, it is clear that striking color differences to determine their maturity.
The evaluation stage of sugarcane detection is done by calculating the precision-recall curve value,
where the precision-recall curve formula
Pricision :
𝑡ℎ𝑒 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐𝑜𝑟𝑟𝑒𝑐𝑡 𝑚𝑎𝑡𝑢𝑟𝑖𝑡𝑦 𝑜𝑓 𝑠𝑢𝑔𝑎𝑟𝑐𝑎𝑛𝑒 𝑑𝑒𝑡𝑒𝑐𝑡𝑖𝑜𝑛 (8)
𝑡ℎ𝑒 number of sugarcane maturity detection

5
3rd Annual Applied Science and Engineering Conference (AASEC 2018) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 434 (2018) 012065 doi:10.1088/1757-899X/434/1/012065
1234567890‘’“”

Recall :
𝑡ℎ𝑒 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑐𝑜𝑟𝑟𝑒𝑐𝑡 𝑚𝑎𝑡𝑢𝑟𝑖𝑡𝑦 𝑜𝑓 𝑠𝑢𝑔𝑎𝑟𝑐𝑎𝑛𝑒 𝑑𝑒𝑡𝑒𝑐𝑡𝑖𝑜𝑛 (9)
𝑡ℎ𝑒 𝑛𝑢𝑚𝑏𝑒𝑟 𝑜𝑓 𝑡𝑟𝑢𝑒 𝑚𝑎𝑡𝑢𝑟𝑖𝑡𝑦 𝑜𝑓 𝑠𝑢𝑔𝑎𝑟𝑐𝑎𝑛𝑒
Here are the results precision and recall from sugarcane maturity detection
Table 2. Precision and recall values of Sugarcane
Method Precision Recall
SVM 87,4% 85,7%
Naïve Bayes 62,3% 89,0%
ANN 86,6% 83,3%
Table 2. showed using a collection of sugarcane datasets can yield that SVM method has the highest
precision value of 87.4%, while Naive Bayes has the highest recall value of 89% in the dataset set.

Figure 3. The classification graph using SVM. Figure 4. The classification graph using Naive
Bayes.

Figure 5. The classification graph using ANN.

Figure 2-4 shows the graphic results of the three classification methods used. Classification using SVM
yields high accuracy values for each class. The naive bayes method produces high frequency values in
the mature and semi-mature classes but has no high accuracy in the immature or raw classes. In the
artificial neural network method, it produces high accuracy values for each class, but if the accuracy
value is compared with the SVM method, then the SVM method has the highest accuracy between the

6
3rd Annual Applied Science and Engineering Conference (AASEC 2018) IOP Publishing
IOP Conf. Series: Materials Science and Engineering 434 (2018) 012065 doi:10.1088/1757-899X/434/1/012065
1234567890‘’“”

three methods. Based result on the trial and analysis above it is known that Support Vector Machine has
the best results with precision and recall of 87,4% and 85,7% respectively.

4. Conclusion
Sugarcane is a long duration crop which reaches its maturity in 11-12 months. Crop growth is very slow
at the initial stage i.e. it takes 25-30 days to complete germination and another 90-95 days to complete
tillering. Knowledge of sugarcane variety is a difficult thing to do. The data on this knowledge is a
complex data. Currently, varieties of sugarcane can be done manually by expert sugarcane. This study
proposes to build maturity stem detection of cane.
Image processing needs a dataset includes several stages produce 300 image cane. Based result on
the trial and analysis above it is known that Support Vector Machine has the best results with precision
and recall of 87,4% and 85,7% respectively. In this study proposes all third algorithm for classification
method to compare which algorithm has the highest accuracy.
This dataset was good enough to be used for classification because not only the dataset affects but
amount of percentage of training data and testing data also affects the classification results.

References
[1] M P R Beulah and O A M Jafar 2017 Prediction of Sugarcane Diseases using Data Mining
Techniques 2016 IEEE Int. Conf. Adv. Comput. Appl. 6(7) pp. 313–318
[2] M Sujaritha, S Annadurai, J Satheeshkumar, S Kowshik Sharan, and L Mahesh 2017 Weed
detecting robot in sugarcane fields using fuzzy real time classifier Comput. Electron. Agric.
134 pp. 160–171
[3] E Santoro, E M Soler and A C Cherri 2017 Route optimization in mechanized sugarcane
harvesting Comput. Electron. Agric.141 pp. 140–146
[4] R V H G Utomo, Adi Heru, and Riyanarto Sarno 2016 Weighted Directed Acyclic Graph
Similarity 2016 Int. Conf. Information, Commun. Technol. Syst. pp. 226–230
[5] S Kumar, S Mishra, P Khanna and Pragyan 2017 Precision Sugarcane Monitoring Using SVM
Classifier Procedia Comput. Sci. 122 pp. 881–887
[6] R K Dewi and R V H Ginardi 2014 Feature extraction for identification of sugarcane rust disease
Proc. 2014 Int. Conf. Information, Commun. Technol. Syst. ICTS 2014 14 pp. 99–103
[7] M P Mello, C Atzberger and A R Formaggio 2014 Near real time yield estimation for sugarcane
in Brazil combining remote sensing and official statistical data Int. Geosci. Remote Sens.
Symp. pp. 5064–5067
[8] M Wang, Y Wan, Z Ye and X Lai 2017 Remote sensing image classification based on the optimal
support vector machine and modified binary coded ant colony optimization algorithm Inf. Sci.
(Ny). 402 pp. 50–68
[9] Z E Rasjid and R Setiawan 2017 Performance Comparison and Optimization of Text Document
Classification using k-NN and Naïve Bayes Classification Techniques Procedia Comput. Sci.
116 pp. 107–112
[10] S Kaymak, A Helwan and D Uzun 2017 Breast cancer image classification using artificial neural
networks Procedia Comput. Sci. 120 pp. 126–131
[11] A Premana, A P Wijaya and M A Soeleman 2017 Image segmentation using Gabor filter and K-
means clustering method 2017 Int. Semin. Appl. Technol. Inf. Commun. pp. 95–99
[12] M P Arakeri and Lakshmana 2016 Computer Vision Based Fruit Grading System for Quality
Evaluation of Tomato in Agriculture industry Procedia Comput. Sci. 79 pp. 426–433
[13] A Younesi and M C Amirani 2017 Gabor Filter and Texture based Features for Palmprint
Recognition Procedia Comput. Sci.108 June pp. 2488–2495