2020 3rd International Conference on Information and Communications Technology (ICOIACT)
Identification of Disease in Potato Leaves
Using Convolutional Neural Network
(CNN) Algorithm
Abdul Jalil Rozaqi
Departement of Informatics engineering
Universitas Amikom Yogyakarta, Indonesia
Email: abdul.13@students.amikom.ac.id
Abstract— Potato crops have many benefits for human
life, one of the most useful benefits of potatoes for
humans is the carbohydrate content in them and
carbohydrates are the main food for humans. The
development of potato crop agriculture is very
important for the sustainability of human life. There
are several obstacles in developing potato farming,
including a disease that attacks potato leaves which if
left untreated will result in poor production or even
crop failure in the future. One of the obstacles in the
development of potato plants is the disease on potato
leaves, namely early blight caused by the fungus
Alternia solani, then late bligt disease caused by
Microbe phytopthora infestans de bary. This disease has
its respective symptoms so that farmers can take
precautions if they see symptoms on potato leaves, but
in this preventive step can only be done by experts who
have knowledge in the field of diseases in potato plants
while the average farmer does not have sufficient
knowledge. So, the identification process becomes less
accurate and takes a long time. Technology in the field
of informatics in the form of digital image processing
can be used to solve problems in disease identification
in potato leaves, so this research will propose the right
method for detecting disease in potato leaves. The
identification process in this study uses three types of
data in the form of healthy leaves, early blight, and late
blight. The method used to identify is deep learning
using the Convolutional Neural Network (CNN)
architecture. The result of this research is that the
70:30 data division produces better accuracy than the
80:20 data division. The accuracy obtained is 97% on
training data and 92% on validation data using 20
batch sizes at 10 epochs.
Keywords—Potato diseases, late blight, early blight,
identification, CNN
Resumen— Los cultivos de papa tienen
muchos beneficios para la vida humana, uno
de los beneficios más útiles de las papas
para los humanos es el contenido de
carbohidratos en ellos y los carbohidratos
son el alimento principal para los humanos.
El desarrollo de la agricultura del cultivo de
papa es muy importante para la
sustentabilidad de la vida humana. Existen
varios obstáculos para desarrollar el cultivo
de la papa, incluida una enfermedad que
ataca las hojas de la papa y que, si no se
trata, dará como resultado una producción
deficiente o incluso la pérdida de la cosecha
en el futuro. Uno de los obstáculos en el
desarrollo de las plantas de patata es la
Authorized licensed use limited to: Carleton University. Downloaded on June 15,2021 at 12:07:06 UTC from IEEE Xplore. Restrictions apply.
Andi Sunyoto
Departement of Informatics engineering
Universitas Amikom Yogyakarta, Indonesia
Email: andi@amikom.ac.id
enfermedad de las hojas de patata, a saber,
el tizón temprano causado por el hongo
Alternia solani, y luego la enfermedad del
tizón tardío causada por el microbio
phytopthora infestans de bary. Esta
enfermedad tiene sus respectivos síntomas
por lo que los agricultores pueden tomar
precauciones si ven síntomas en las hojas
de papa, pero en este paso preventivo solo
lo pueden hacer expertos que tengan
conocimientos en el campo de las
enfermedades en las plantas de papa
mientras que el agricultor promedio no
tiene los suficientes conocimiento. Por lo
tanto, el proceso de identificación se vuelve
menos preciso y lleva mucho tiempo. La
tecnología en el campo de la informática en
forma de procesamiento digital de
imágenes se puede utilizar para resolver
problemas en la identificación de
enfermedades en hojas de papa, por lo que
esta investigación propondrá el método
adecuado para detectar enfermedades en
hojas de papa. El proceso de identificación
en este estudio utiliza tres tipos de datos en
forma de hojas sanas, tizón temprano y
tizón tardío. El método utilizado para
identificar es el aprendizaje profundo
utilizando la arquitectura de red neuronal
convolucional (CNN). El resultado de esta
investigación es que la división de datos
70:30 produce una mayor precisión que la
división de datos 80:20. La precisión
obtenida es del 97 % en datos de
entrenamiento y del 92 % en datos de
validación utilizando 20 tamaños de lote en
10 épocas.
I. INTRODUCTION
Potato plants have many benefits for human life.
The main human need is carbohydrates and potatoes
are one of the foods that contain carbohydrates. But
in the development of potato crops there are
obstacles that must be faced by farmers, this obstacle
is in the form of potato leaf disease. If not treated
 2020 3rd International Conference on Information and Communications Technology (ICOIACT)
promptly, the potato leaf disease will result in a
decrease in potato farm income, and this means a
decrease in food production [1]. So it is necessary to
detect disease in plants at the right time in order to
effectively control and prevent plant diseases. [2].
2020 3rd International Conference on Information and Communications Technology (ICOIACT) | 978-1-7281-7356-6/20/$31.00 ©2020 IEEE | DOI: 10.1109/ICOIACT50329.2020.9332037
The most common
978-1-7281-7356-6/20/$31.00 ©2020 IEEE
Authorized licensed use limited to: Carleton University. Downloaded on June 15,2021 at 12:07:06 UTC from IEEE Xplore. Restrictions apply.
diseases in potato plants are diseases that attack the
leaves of potato plants, namely early blight and late
blight. Cold and humid places are one of the factors
for leaf disease in potato plants [3].
A potato leaf disease called early blight has early
symptoms characterized by circular spots on the
middle of the leaves and it could also be on the edges
of the leaves as shown in Fig. 1 (a). Then these spots
will widen and the color of the leaves turns brown, the
fungus Alternia solani is the cause of this leaf disease.
Furthermore, Microbe Phytopthora infestans de Bary
is the main cause of potato leaf disease called late
blight, plant leaves affected by this disease can cause
plant damage. The Fig. 1 (b) shows a leaf with late
blight, marked by the appearance of black lesions on
the leaves and will continue to propagate [4].
(a) (b)
Fig. 1. Potato disease (a) early blight (b) late blight
This identification can help agricultural managers
to provide effective and efficient handling of plants
that are not healthy or abnormal. With the
development of technology today there have been
many digital image studies in agriculture both to
identify diseases or identify good agricultural
production. One of these digital image studies is to
identify leaf rot in potato plants.
This study aims to create a system that can help
farmers or agricultural managers in identifying
diseases in potato leaves by using data on potato leaf
images. The identification of leaves in potato plants is
divided into three parts, namely potato plants with
healthy leaves, late blight, and early blight. So in this
study will identify this using the Convolutional Neural
Network (CNN) architecture which is one of the Deep
Learning methods. The data used in the form of
72
 2020 3rd International Conference on Information and Communications Technology (ICOIACT)
disease data on leaves of potato plants obtained from
the website Kaggle with the name PlantVillage [5].
II. RELATED LITERATURE
Previous studies that have conducted trials in
detecting diseases in plants include the following:
Suttapakti in his research titled "Potato Leaf
Disease Classification Based on Distinct Color and
Texture Feature Extraction" in identifying leaf
diseases in potato plants focuses on feature extraction,
because according to him in doing classification will
get good results depending on accurate extraction
features [6]. He was comparing the color extraction
feature between Color Moments (CM) and Boundary
Color (BC) and Maximum-minimum Color
Difference (MCD) on 300 potato leaf images and
results that MCD has better accuracy than others with
a level of 82.5%.
Islam. M in his research "Detection of Potato
Diseases Using Image Segmentation and Multiclass
Support Vector Machine" uses the Support Vector
Machine (SVM) method to detect disease in potato
leaves with the amount of leaf image data used 200
data of diseased leaves and 100 normal or healthy
leaves. The process in this study is to segment the
image to display only leaf disease without displaying
the background and normal leaves in the sense of leaf
green, then the leaf image will be extracted using the
Gray Level Co-occurrence Matrix (GLCM). The
results of this study have an accuracy of 95% [7].
Research conducted by Mim "Leaves Diseases
Detection of Tomato Using Image Processing"
conducted a classification with 6 classes in the form of
healthy, late blight, yellow curved, tomato mosaic,
bacterial spots, septorial leaf spots. The algorithm
used for classification is Convolutional Neural
Network (CNN). In this study the best accuracy
results obtained by running 30 epochs with 92.61%
accuracy training results and validation accuracy
produces 96.55% [8].
Prakash in his research entitled "Detection of Leaf
Diseases and Classification using Digital Image
Processing" segmented to remove background in leaf
images using K-Means clustering, and also extracted
leaf textures using the Gray Level Co-Occurrence
Matrix (GLCM) feature so that showing only leaves
that are not green or that are not normal. In the final
stage, classification using the Support Vector Machine
(SVM) algorithm and produces an accuracy of 90% of
the image data used as many as 60 images [9].
From the description of the research that has been
done in identifying plant diseases using leaf image
datasets, it can be seen that there are those using the
Support Vector Machine (SVM) algorithm with
additional features extracting leaf texture features and
producing good accuracy [7]. In the research
conducted by Mim is to detect diseases of tomatoes
with Convolutional Neural Network (CNN) and
provide excellent accuracy in 6 classes [8]. So, in this
Authorized licensed use limited to: Carleton University. Downloaded on June 15,2021 at 12:07:06 UTC from IEEE Xplore. Restrictions apply.
study will identify diseases in potato plants using the
CNN architecture from Deep Learning Method.
Los estudios previos que han realizado ensayos para
detectar enfermedades en las plantas incluyen los
siguientes: Suttapakti en su investigación titulada
"Clasificación de enfermedades de la hoja de papa
basada en la extracción de características de textura y
color distintivo" en la identificación de enfermedades de
las hojas en las plantas de papa se enfoca en la
extracción de características, porque según él, al hacer la
clasificación se obtendrán buenos resultados
dependiendo de las características de extracción precisas
[6] . Estaba comparando la función de extracción de
color entre los momentos de color (CM) y el color de
contorno (BC) y la diferencia de color máxima-mínima
(MCD) en 300 imágenes de hojas de patata y resultó que
MCD tiene una mayor precisión que otras con un nivel
de 82,5 %. Islam. M en su investigación "Detección de
enfermedades de la papa usando segmentación de
imágenes y máquina de vectores de soporte multiclase"
utiliza el método de máquina de vectores de soporte
(SVM) para detectar enfermedades en hojas de papa con
la cantidad de datos de imagen de hoja utilizados 200
datos de hojas enfermas y 100 datos normales o hojas
sanas. El proceso en este estudio es segmentar la imagen
para mostrar solo la enfermedad de la hoja sin mostrar el
fondo y las hojas normales en el sentido de hoja verde,
luego la imagen de la hoja se extraerá usando la matriz
de co-ocurrencia de nivel gris (GLCM). Los resultados
de este estudio tienen una precisión del 95% [7]. La
investigación realizada por Mim "Detección de
enfermedades de las hojas del tomate mediante el
procesamiento de imágenes" realizó una clasificación
con 6 clases en forma de saludable, tizón tardío, amarillo
curvo, mosaico del tomate, manchas bacterianas,
manchas foliares septoriales. El algoritmo utilizado para
la clasificación es Red Neural Convolucional (CNN). En
este estudio, los mejores resultados de precisión se
obtuvieron al ejecutar 30 épocas con un 92,61 % de
precisión en los resultados del entrenamiento y la
precisión de validación produce un 96,55 % [8].
Prakash, en su investigación titulada "Detección de
enfermedades de las hojas y clasificación mediante el
procesamiento de imágenes digitales", segmentó para
eliminar el fondo de las imágenes de las hojas mediante
el agrupamiento de K-Means, y también extrajo las
texturas de las hojas mediante la función Matriz de co-
ocurrencia de nivel de gris (GLCM) para que mostrar
solo hojas que no son verdes o que no son normales. En
la etapa final, la clasificación utiliza el algoritmo
Support Vector Machine (SVM) y produce una precisión
del 90 % de los datos de imagen utilizados hasta 60
imágenes [9]. A partir de la descripción de la
investigación que se ha realizado para identificar
enfermedades de las plantas usando conjuntos de datos
de imágenes de hojas, se puede ver que hay aquellos que
usan el algoritmo Support Vector Machine (SVM) con
características adicionales que extraen características de
la textura de la hoja y producen una buena precisión [7] .
En la investigación realizada por Mim se trata de
detectar enfermedades de tomates con Red Neural
Convolucional (CNN) y brindar una excelente precisión
en 6 clases [8]. Entonces, en este estudio identificará
enfermedades en plantas de papa utilizando la
arquitectura CNN de Deep Learning Method
 2020 3rd International Conference on Information and Communications Technology (ICOIACT)

Authorized licensed use limited to: Carleton University. Downloaded on June 15,2021 at 12:07:06 UTC from IEEE Xplore. Restrictions apply.
 2020 3rd International Conference on Information and Communications Technology (ICOIACT)

Dataset Collection

Preprocessing Data

Classification

Evaluation

III. PROPOSED METHOD

In this paper, there are several stages of research

completion as shown in Fig 1 in the form of a research
framework. In this research framework, there are four
stages in the form of dataset collection, image data
pre-processing, training data, data evaluation.

Fig. 2. Research Framework

Explanation of each discussion about Fig 1 will be

discussed further:
A. Dataset Collection
The dataset used in this study is a picture of the
leaves of a potato plant divided into three classes: the
healthy leaves shown in Figure 2, early blight has
shown in Figure 3, and the late blight shown in
Figure 3. This dataset was obtained from the Kaggle
website under the name "PlantVillage Dataset"
uploaded by Tairu Oluwafemi Emmanuel and the
latest changes were made in October 2018 [5]. The
amount of data used is 500 data late blight, 500 initial
early blight data, and 150 healthy leaf data. Details of
the data agreed upon in Table 1.
TABLE 1 DETAIL DATASET
Sampels Number Repository
Late Blight 500 Kaggle
(PlantVillage)
Early Blight 500
Leaf Helathy 150
Total 1125
All images used in this study will be resized to
150x150 to speed up processing. Fig 3 are examples
of pictures from each data class used Fig 3 (a) leaf
healthy, Fig 3 (b) early blight and Fig 3 (c) late
blight.

73

Authorized licensed use limited to: Carleton University. Downloaded on June 15,2021 at 12:07:06 UTC from IEEE Xplore. Restrictions apply.
 2020 3rd International Conference on Information and Communications Technology (ICOIACT)

convolution process, matrix multiplication will be

carried out on the filter and leaf image area. As
shown in Fig. 4 it can be seen that this convolution
process multiplies the pixels in the image by the filter
pixels. This research will use 4 layers of convolution.
D. Pooling
After the convolution carried out, then it will do
(a) pooling. Pooling that often used is MaxPooling.
Pooling here means the process carried out to get
images with smaller pixels but still by maintaining
the information in the image. The pooling process can
be seen in Fig 5, where in the image area with a
particular pixel area will be done pooling by selecting
one of the highest pixels. This process is constructive
because it will reduce the size of each image and will
(b) (c) be able to speed up the classification process.
Fig. 3 data (a) leaf healthy (b) early
blight (c) late blight
B. Preprocessing Data
At this stage the amount of data used is 1150
images from 3 classes as in Figure 5, the class
divisions are late blight, early blight and leaf healthy.
In Table 2, the details of the distribution of each data
from each class used are divided into training data
and testing data with data division of 80:20 and Fig. 5 MaxPooling
70:30. The results of each of these data sharing will
be compared to the results of its accuracy to E. Classification
determine which is better in dividing the proportion The next step is to classify images using the
of data. The leaf image used first will be resized to Convolutional Neural Network (CNN) architecture
150x150 to speed up classification processing. from Deep Learning Method. Convolutional Neural
TABLE 2 DETAIL OF DATASET Network (CNN) architecture from Deep Learning
80:20 70:30 Method is included in the supervised learning method
Dataset Train Val Train Val
where identification of an image by training existing
image data and targeting image variables.
Late Blight 400 100 350 150
The convolutional layer in the Convolutional
Early Blight 400 100 350 150 Neural Network (CNN) architecture helps neural
Leaf Healthy 122 30 105 45 networks in the CNN method recognize potato leaves
Total 922 230 805 345
based on the attributes they have. The neural network
can recognize images of potato leaves based on the
The distribution of data used is division 80:20 and pixels in the picture
70:30, for data deviding 90:10 is not used in this This research will use an image with a size of
study because considering the 150 leaf helathy data, 150x150x3, which means here is a 150x150 size
if the data validation data used helati leaves only 10% image and this image has three channels, namely red,
it will not sufficient for the validation process with green, and blue (RGB). This leaf image will be
the number of batch sizes used in this study is 20 convoluted first with a filter. Then pooling will be
batch sizes. done to reduce the image resolution while
maintaining image quality, pooling used is
C. Convolution MaxPooling on the input image.
The next process is a fully connected layer,
wherein this process is doing flatten. The purpose of
flattening here is to change the feature map resulting
from pooling into vector form. For more details,
shown in Fig 6 architecture of the Convolutional
Neural Network (CNN).
Fig. 4 Convolution process
Convolutional uses filters to recognize the
attributes present in the leaf image. In this

74

Authorized licensed use limited to: Carleton University. Downloaded on June 15,2021 at 12:07:06 UTC from IEEE Xplore. Restrictions apply.
 2020 3rd International Conference on Information and Communications Technology (ICOIACT)

TABLE 4 RESULT FROM FIT MODEL 70:30

Data Training Data Testing
Epoch
Acc Loss Val Acc Val Loss
1 0.4611 0.6103 0.4364 0.5946
2 0.5758 0.5156 0.7364 0.3630
3 0.6930 0.3964 0.8773 0.2759
Fig. 6 Convolutional Neural Network Architecture 8 0.9478 0.0898 0.8636 0.2510
9 0.9605 0.0666 0.9000 0.1728
This research has the proposed model for
Convolutional Neural Network (CNN) architecture in 10 0.9758 0.0560 0.9227 0.1207
identifying diseases in potato leaves is shown in
Table 3 using 4 convolution layers and 4 maxPooling. Table 4 that the results of the classification on the
train data are divided into 70:30. The results of the
TABLE 3 CONVOLUTIONAL NEURAL NETWORK MODEL
first epoch display the accuracy value on the train data
Layer Output Shape Param is 46% with a loss value obtained that is 61% and so
conv2d (Conv2D) (None, 148, 148, 32) 896 on until the 10th epoch displays an accuracy value of
97% with a loss value obtained at 0.05%. Where as in
max_pooling2d
(None, 74, 74, 32) 0 the validation data, the accuracy value on the first
(MaxPooling2D)
epoch is 43% with a loss value obtained by 59% and
conv2d_1 (Conv2D) (None, 72, 72, 64) 18496 so on until the 10th epoch displays an accuracy value
max_pooling2d_1 of 92% with the loss value obtained is 12%.
(None, 36, 36, 64) 0
(MaxPooling2D)

conv2d_2 (Conv2D) (None, 34, 34, 128) 73856

max_pooling2d_2
(None, 17, 17, 128) 0
(MaxPooling2D)

conv2d_3 (Conv2D) (None, 15, 15, 256) 295168

max_pooling2d_3
(None, 7, 7, 256) 0 (a) (b)
(MaxPooling2D)
flatten (Flatten) (None, 12544) 0 Fig. 7 Training and validation from 70:30 data
dividing (a) accuracy (b) loss
dense (Dense) (None, 512) 6423040 The graph shown in Figure 7 (a) is the accuracy
graph and the graph shown in Figure 7 (b) is the loss
dense_1 (Dense) (None, 3) 1539
graph for the 70:30 data division. The blue line shows
Total params: 6,812,995 training data, and the orange line shows validation
Trainable params: 6,812,995 data. This graph shows that the fit model made is good
Non-trainable params: 0 because the increased accuracy and decreased loss on
each epoch are stable.
IV. RESULT AND EVALUATION TABLE 5 RESULT FROM FIT MODEL 80:30
Epoch is a training process on neural networks Data Training Data Testing
Epoch
until it returns to the initial stage in one round when Acc Loss Val Acc Val Loss
all datasets go through this process. In training data 1 0.6077 0.4364 0.5796
0.4412
with a neural network model, if you only use one
2 0.6522 0.4482 0.7909 0.3902
epoch, this will be too large and will stress the
training process in the dataset, because the data used 3 0.8005 0.3083 0.8091 0.3253
is quite a lot, it is necessary to divide the data rate per 8 0.9258 0.1222 0.9182 0.1352
batch (batch size). In this research, there were 20
9 0.9629 0.0610 0.9636 0.0861
batch sizes and to determine the number of epochs,
the researchers adjusted the number of batch sizes to 10 0.9655 0.0623 0.9273 0.1755
the number of samples used.
The next step is to conduct training on the potato Then in Table 5 shows the results of the
leaf image which has been divided by the fit model. classification on the data are devided into 80:20. The
Table 4 is the result of the fit model in the 70:30 data results of the first epoch display the accuracy value on
the train data is 44% with a loss value obtained that is
division, then Table 5 shows the results of the fit
60% and so on until the 10th epoch displays an
model in the 80:20 data division. It can be seen from accuracy value of 96% with a loss value obtained at
epoch 1 to epoch 10 that the accuracy value on the 0.06%. Where as in the testing data, the accuracy
train data and the accuracy value on the validation value on the first epoch is 43% with a loss value
data has increased. obtained by 57% and so on until the 10th epoch

75

Authorized licensed use limited to: Carleton University. Downloaded on June 15,2021 at 12:07:06 UTC from IEEE Xplore. Restrictions apply.
 2020 3rd International Conference on Information and Communications Technology (ICOIACT)

displays an accuracy value of 92% with the loss value

obtained is 17%. [2] J. Chen, J. Chen, D. Zhang, Y. Sun, and Y. A.
Nanehkaran, “Using deep transfer learning
for image-based plant disease identification,”
Comput. Electron. Agric., vol. 173, no.
March, p. 105393, 2020, doi:
10.1016/j.compag.2020.105393.
[3] A. Fitriana and L. Hakim, “Endophytic
Fungus effectiveness Origin Root Crops
(a)
(b) Potatoes in Compost Media Skin Diseases
Fig. 8 Training and validation from 80:20 data Coffee to Suppress Development Potato Leaf
dividing (a) accuracy (b) loss blight (Phytophtora infestans),” J. Agrista,
The graph is shown in Fig 8 (a) is graph of vol. 23, no. 1, pp. 9–14, 2019.
accuracy and graph is shown in Fig 8 (b) is graph of
loss for the 80:20 data division. This graph also shows [4] P. Patil, N. Yaligar, and S. . Meena,
that the fit model made with 80:20 data dividing is “Comparision of Performance of Classifiers -
good because the increased accuracy and decreased SVM, RF and ANN in Potato Blight Disease
loss for each epoch are as stable. Detection Using Leaf Images,” IEEE Int.
From this research it can be concluded that based Conf. Comput. Intell. Comput. Res. ICCIC
on the implementation of the model, the distribution 2017, pp. 1–5, 2017, doi:
of the dataset has little effect on the results of 10.1109/ICCIC.2017.8524301.
accuracy, from the results carried out on potato leaf [5] T. O. Emmanuel, “PlantVillage Dataset From
data has a slightly better result by dividing the 70:30 Kaggle,” kaggle.com, 2018.
dataset compared to dividing the dataset at 80:20. The https://www.kaggle.com/emmarex/plantdisea
image used is changed to 150x150 size then se (accessed Jul. 24, 2020).
classification is done with 10 epochs with a batch size
of 20 with a total data of 1150 images. The accuracy [6] U. Suttapakti and A. Bunpeng, “Potato Leaf
value for the training data is 97% and for the Disease Classification Based on Distinct
validation accuracy is 92% for the 70:30 data Color and Texture Feature Extraction,” Proc.
dividing, while the accuracy value for the training data - 2019 19th Int. Symp. Commun. Inf. Technol.
is 96% and for the validation accuracy is 92% for the Isc. 2019, no. Mcd, pp. 82–85, 2019, doi:
80:20 data dividing. 10.1109/ISCIT.2019.8905128.
V. CONCLUSION [7] M. Islam, A. Dinh, K. Wahid, and P.
From this research it can be concluded that based Bhowmik, “Detection of potato diseases
on the implementation of the model carried out, the using image segmentation and multiclass
distribution of the dataset has an effect on the results support vector machine,” Can. Conf. Electr.
of accuracy, from the results carried out on potato Comput. Eng., pp. 8–11, 2017, doi:
leaf data has good results by dividing the dataset 70% 10.1109/CCECE.2017.7946594.
and 30% and the image used is changed in size [8] T. T. Mim, M. H. Sheikh, R. A. Shampa, M.
150x150. On the 10th epoch with a batch size of 20 S. Reza, and M. S. Islam, “Leaves Diseases
with a total training data of 805 images and testing Detection of Tomato Using Image
data of 345 images, the accuracy value on the train Processing,” in Proceedings of the SMART,
data is 97% and for validation accuracy is 92%. 2019, pp. 244–249, doi:
REFERENCES 10.1109/smart46866.2019.9117437.

[1] B. Kumbar et al., “Field application of [9] R. Meena Prakash, G. P. Saraswathy, G.

Bacillus subtilis isolates for controlling late
blight disease of potato caused by
Phytophthora infestans,” Biocatal. Agric.
Biotechnol., vol. 22, no. April, p. 101366,
2019, doi: 10.1016/j.bcab.2019.101366.
Ramalakshmi, K. H. Mangaleswari, and T.
Kaviya, “Detection of leaf diseases and
classification using digital image processing,”
Int. Conf. Innov. Information, Embed.
Commun. Syst., pp. 1–4, 2017, doi:
10.1109/ICIIECS.2017.8275915.

76

Authorized licensed use limited to: Carleton University. Downloaded on June 15,2021 at 12:07:06 UTC from IEEE Xplore. Restrictions apply.