Classification of sugarcane diseases using machine learning algorithms

Jannie Leila N. Vergara, Edwin R. Arboleda*)

1)
Department of Computer and Electronics Engineering, College of Engineering and Information Technology,
Cavite State University
Indang, Cavite, Philippines 4122

How to cite: E. R. Arboleda and J. L. N. Vergara, " Classification of sugarcane diseases using machine learning
algorithms," Jurnal Teknologi dan Sistem Komputer, vol. x, no. x, pp. xx-xx, 202x. doi:
10.14710/jtsiskom.x.x.202x.xx-xx, [Online].

Abstract - Sugarcane (Saccharum officinarum L.) many parts of Asia due to an abundance of insect pests
belongs to the Poaceae (grass) family's and diseases [6]. Sugarcane is contaminated with over
Andropogoneae tribe. Modern sugarcane cultivars are 200 diseases caused by fungi, viruses, bacteria,
primarily the product of interspecific hybridizations phytoplasmas, and nematodes, which are among the
between Saccharum species. Sugarcane yields have leading causes of sugarcane yield losses worldwide.
been rigorously abridged in various parts of Asia as Viruses are among the world's most essential sugarcane
attacked by number of diseases. Amongst all the diseases, both historically and in the future, causing
diseases, red rot and smut are causing significant large epidemics and significant losses. As a result,
outbreaks in recent years, causing 30-100% yield loss sugarcane viruses must still be recognized in breeding
in commercial sugarcane cultivars. The MATLAB programs worldwide, where the collection and removal
2021 Classification Learner App having 28 different of susceptible clones after resistance tests have become
machine learning was used in classifying the different standard practice [7].
diseases. The study came up to a Medium Neural Sugarcane crop is attacked by a massive number of
Network classifier. This model obtained an accuracy insect pests and diseases [8]. A large number of insect
of 100% with a training time of 0.85113 seconds. pests and diseases attack the sugarcane crop. Complexes
Keywords – sugarcane; medium neural network, of branch feeders, sap-sucking insects (e.g., aphids,
classification learner application thrips, mealybugs), root feeders (e.g., white grubs, stem
borers), and spider mites are among the arthropod pests
associated with the crop across the world [9].
I. INTRODUCTION
Being a long durational crop, sugarcane is prone to
Sugarcane (Saccharum officinarum L.) belongs to several diseases caused by pathogens, viz. fungi,
the Poaceae (grass) family's Andropogoneae tribe. bacteria, viruses, and phytoplasmas-like organisms.
Modern sugarcane cultivars are primarily the product of Amongst all the diseases, red rot and smut are causing
interspecific hybridizations between Saccharum species significant outbreaks in recent years, causing 30-100%
[1]. Saccharum is a complex genus with a high yield loss in commercial sugarcane cultivars throughout
polyploidy rate and regular aneuploidy [2]. Sugarcane India [10].
is the world's largest crop, with more than 1800 mt Generally, this research aims to classify sugarcane
generated in 2017 on a surface of nearly 26 mt. diseases with the help of MATLAB Classification
Sugarcane is infected by a range of viruses all over the Learner App.
world, owing to its vegetative propagation [3].
Sugarcane is one of the top ten most commonly II. METHODOLOGY
planted crops on the planet. Globally, about one billion
tons are harvested per year [4]. Sugarcane diseases are A. Classification
either seed or soil-borne, making it nearly difficult to
contain the disease once it has spread through the field, The researcher used MATLAB Classification
except with agrochemicals. One or more preventive Learner App and it contains 28 machine learning
mechanisms may be used to reduce disease incidence. algorithms for classifying the samples. The App trains
For the control of sugarcane diseases, no single method the data and validates the classification in terms of
has proven to be reliable. The most successful method accuracy and training speed. The data were inputted to
for managing all diseases is integrated sugarcane disease the Classification Learner App using 5 folds, this means
management. Agronomic, cultural, chemical, and that 180 were used for training and the 36 were used for
biological control steps are all included [5]. testing and validation [11].
.
Sugarcane yields have been severely reduced in
*)
Correspondence author (Edwin R. Arboleda)
Email: [email protected]
@cvsu.edu.ph
B. Image Processing Prolonged
storage or
To make some predictions in the data in a large set exposure to
of information, the image processing technique will
high relative
study and construct algorithms. Hence Image Processing 1 REDROT humidity,
is processing the images with a different form of environmenta
mathematical functions and operations by some signal l pollution,
processing. It is usually referring to digital image and high
processing to acquits the images for producing the input temperature
images. The digital image used in this research is a two-
dimensional image that has values by pixels and Multicellular
pictures. So by this image processing, similar sets of fungi
plant leaf diseases are identified from various plants. WHIP characterized by
2
SMUT their large
III. RESULTS AND DISCUSSION numbers of
teliospores
A. Description of Dataset
Affect the
The datasets are collected from Kaggle [11]. The vascular
3 WILT
sugarcane data frame has 180 rows and 6 columns as system of
shown in Table 1. It represents a randomized block plants.
design with 45 varieties of sugarcane and 4 blocks as
shown in Table 3. Yellow bud
mosaic in
Table I. Dataset Description peaches,
No. of No. of yellow vein
Datasets
Attributes Instances 4 RINGSPOT in grapes,
Sugarcane and stunted
6 180 growth in
Datasets
gladiolus and
Narcissus.
Table II. Sugarcane Dataset Attributes
Sample No. Attributes ID Different
genotypes in
1 S S0
one
2 N S1
individual,
3 R S2
5 MOSAIC who has
4 X S3
developed
5 Var S4
from a single
6 Block S5
fertilized
egg.
Table III. Sugarcane Dataset Attributes The presence
Sample No. Attributes ID of pin head
1 S S0 like orange-
2 N S1 colored dots
RATOON
3 R S2 6 of bacteria
STUNT
4 X S3 on the
internal soft
tissue in the
Where: nodal region.
R – No. of diseased shoots
X – No. of pieces of the stem
Var – Indicates variety of sugarcane in each plot Table 5 shows the classification results. The
Block – Factor for the blocks classifier used are the 28 machine learning algorithms of
MATLAB Classification Learner App. There are 45
varieties of sugarcane in each block, and the chosen
Table IV. Image Processing [12] settings are five folds. Five Folds is the number of
samples per class; 45 divided by 5 is equal to 9. A total
Name of the
S. No Images Description of 45 samples per class was used as training samples,
disease
and nine samples were used as test samples.
 Table V. Classification Results Figure 1 shows the scatter plot of the Medium
Accuracy Training Neural Network for the sugarcane disease. The violet
Classifier Type colors represent Block A, the yellow colors represent
(%) Time (S)
Fine Tree 98.3 5.6625 Block B, the red colors represent Block C, and the blue
Medium Tree 98.3 2.4572 colors represent Block D. It can be seen in Figure 1 that
Coarse Tree 98.3 2.0329 values of the four groups are separable in the viewpoint
Linear Discriminant 100 2.4833 of the Medium Neural Network.
Gaussian Naïve Bayes 93.9 2.028
Kernel Naïve Bayes 92.9 2.5349
Linear SVM 100 3.2829
Quadratic SVM 100 2.1459
Cubic SVM 100 2.0904
Fine Gaussian SVM 74.4 1.8858
Medium Gaussian SVM 94.4 1.8058
Coarse Gaussian SVM 99.4 1.5785
Fine KNN 80 3.9228
Medium KNN 71.1 2.8147
Coarse KNN 48.3 2.8183
Cosine KNN 58.9 3.0026
Cubic KNN 64.4 1.6503
Weighted KNN 82.8 4.4522
Boosted Trees 25 3.5517
Bagged Trees 98.3 4.793
Subspace Discriminant 86.7 2.8469
Subspace KNN 96.7 2.6402
RUS Boosted Trees 25 0.9279
Narrow Neural Network 98.3 1.4661
Medium Neural Network 100 0.85113
Wide Neural Network 100 2.8794 Figure 2. Scatter Plot Confusion Matrix of the Medium
Bi-layered Neural Neural Network
98.9 4.7855
Network
Tri-layered Neural It can be seen in Figure 2, the confusion matrix that
97.8 1.9022 the Medium Neural Network Classifier was able to
Network
correctly identify all of the blocks by 100%.

Figure 3 shows the ROC or the receiver operating

Table 5 also shows that the Linear Discriminant,
characteristics of the Medium Neural Network.
Linear SVM, Quadratic SVM, Cubic SVM, Medium
Neural Network, and Wide Neural Network all achieved
the highest accuracy of 100%, however in terms of
training time the fastest was achieved by the Medium
Neural Network with 0.85113 seconds, compared to the
others who got higher training time in seconds.

Figure 3. ROC Curve of the Medium Neural Network

Figure 1. Scatter Plot of the Medium Neural Network Figure 3 shows that the Receiver Operating
Characteristics curves depicts the relationship between
the classifier and the inputted features. Also, it shows
that the area under the curve (AUC) is 1.00. Since the Implications. Elsevier Ltd., 2020.
value approached 1, the accuracy of the used classifier [4] M. A. Godshall, “Sugar and Other Sweeteners,”
is high. Kent Riegel’s Handb. Ind. Chem. Biotechnol.,
pp. 1657–1693, 2007, doi: 10.1007/978-0-387-
27843-8_35.
[5] A. K. Christian-albrechts-universit, N. P. View,
and A. Zia, “Integrated Control Strategies For
Sugarcane Disease Integrated Control Strategies
For Sugarcane,” Pak Sugar J., no. January
2010, pp. 1–4, 2014.
[6] M. Usman, N. Akram, S. Imtiaz, and ...,
“Sugarcane: Diseases Due to Pests, Pest
Management Strategies and Factors Influencing
the Production of Sugarcane,” Am. Sci. Res. J.
Eng. Technol. Sci., pp. 126–139, 2020, [Online].
Available:
https://www.asrjetsjournal.org/index.php/Ameri
can_Scientific_Journal/article/view/5497.
Figure 4. Parallel Coordinates of the Medium Neural [7] M. Gonçalves, L. Pinto, S. Souza, and M.
Network Landell, “Virus Diseases Of Sugarcane. A
Constant Challenge to Sugarcane Breeding In
Figure 4 shows the parallel coordinates plot of the Brazil,” Funct. Plant Sci. Biotechnol., vol. 6, no.
Medium Neural Network. It can be seen in that the 2, pp. 108–116, 2012.
values were overlapping to each other. Although the [8] R. L. Yadav, S. K. Shukla, A. Suman, and P. N.
values were overlapping, the Medium Neural Network Singh, “Trichoderma inoculation and trash
was able to differentiate the blocks. management effects on soil microbial biomass,
soil respiration, nutrient uptake and yield of
CONCLUSION ratoon sugarcane under subtropical conditions,”
Biol. Fertil. Soils, vol. 45, no. 5, pp. 461–468,
The automated detection of sugarcane leaf diseases 2009, doi: 10.1007/s00374-009-0352-4.
is investigated in this research. The MATLAB 2021 [9] G. Dittrich, D. E. Conlong, and A. Mitchell,
Classification Learner App containing 28 different “Molecular Identification of South African
machine learning algorithms was used in classifying the Sugarcane White Grubs ( Coleoptera :
dataset. The Medium Neural Network Classifier attained Scarabaeidae ),” Seven, no. November 2014, pp.
the highest accuracy validation of 100%, having 264–268, 2006.
0.85113 seconds training time. The researcher analyzed [10] S. Chaudhary, Piyush, Chaudhari, Anand K.,
the efficiency of machine learning techniques based on Cheeran, Dr. A. N., and Goda, “Color
time measurements and their ability to predict sugarcane Transform Based Approach for Disease Spot
diseases. The output of various machine learning Detection on Plant Leaf,” Int. J. Comput. Sci.
techniques can be studied in the future to predict various Telecommun., vol. 3, no. 6, 2012.
crop diseases. [11] J. M. Ashfaque, “Sugarcane Dataset.” Margate,
England, United Kingdom, [Online]. Available:
REFERENCES https://www.kaggle.com/ukveteran/sugarcane-
disease-data.
[1] Norman Simmonds, Sugarcanes. Evolution of
[12] R. Ravikumar, “Prediction of Leaf Diseases by
Crop Plants. 1995. using Machine Learning Techniques-A New
[2] G. Bremer, “Problems in breeding and cytology
Approach to Applied Informatics,” Int. J.
of sugar cane - II. The sugar cane breeding from
Recent Innov. Trends Comput. Commun. J.
a cytological view-point,” Euphytica, vol. 10,
Recent Innov. Trends Comput. Commun., no.
no. 2, pp. 121–133, 1961, doi:
June, pp. 1154–1158, 2017, doi:
10.1007/BF00022203.
10.17762/ijritcc2321-8169.150254
[3] B. L. Patil, Plant Viral Diseases: Economic