Peanut leaf spot disease identification using pre-trained deep convolutional neural network

International Journal of Electrical and Computer Engineering (IJECE)
Vol. 12, No. 3, June 2022, pp. 3005~3012
ISSN: 2088-8708, DOI: 10.11591/ijece.v12i3.pp3005-3012  3005
Journal homepage: http://ijece.iaescore.com
Peanut leaf spot disease identification using pre-trained deep
convolutional neural network
Urbano B. Patayon1
, Renato V. Crisostomo2
1
Computer Science Department, Jose Rizal Memorial State University, Zamboanga del Norte, Philippines
2
Computer Science Department, Mindanao State University–Iligan Institute of Technology, Lanao del Norte, Philippines
Article Info ABSTRACT
Article history:
Received Mar 27, 2021
Revised Dec 24, 2021
Accepted Jan 19, 2022
Reduction of quality and quantity of agricultural products, particularly
peanut or groundnut, is usually associated with disease. This could be solved
through automatic identification and diagnoses using deep learning.
However, this technology is not yet explored and examined in the case of
peanut leaf spot disease due to some aspects, such as the availability of
sufficient data to be used for training and testing the model. This study is
intended to explore the use of pre-trained visual geometry group–16
(VGG16), visual geometry group–19 (VGG19), InceptionV3, MobileNet,
DenseNet, Xception, InceptionResNetV2, and ResNet50 architectures and
deep learning optimizers such as stochastic gradient descent (SGD) with
Momentum, adaptive moment estimation (Adam), root mean square
propagation (RMSProp), and adaptive gradient algorithm (Adagrad) in
creating a model that can identify leaf spot disease by using a total of 1,000
images of leaves captured using a mobile camera. Confusion matrix was
used to assess the accuracy and precision of the results. The result of the
study shows that DenseNet-169 trained using SGD with momentum, Adam,
and RMSProp attained the highest accuracy of 98%, while DenseNet-169
trained using RMSProp achieved the highest precision of 98% among
pre-trained deep convolutional neural network architectures. Furthermore,
this result could be beneficial in agricultural automation and disease
identification systems for peanut or groundnut plants.
Keywords:
Artificial intelligence
Deep convolutional neural
network
Deep learning
Deep learning optimizers
Peanut leaf spot
Transfer learning
This is an open access article under the CC BY-SA license.
Corresponding Author:
Urbano B. Patayon
Computer Science Department, Jose Rizal Memorial State University
Zamboanga del Norte, Philippines
Email: urbanopatayon@jrmsu.edu.ph
1. INTRODUCTION
Peanut (Arachis hypogea L.) is one of the commercially important legumes cultivated by local
farmers in the Philippines [1]. However, low production is associated with low yielding variety, lack of high-
yielding adaptive cultivars, pests and diseases, poor agronomic practices, climate change, and limited use of
inputs [2]. Considering peanut diseases, the most common that affect the production are sclerotina blight, late
leaf spot, northern root-knot nematode, spottedwilt, stem rot, early leaf spot, rust, web blotch, diplodia collar
rot, funky or irregular leaf spot, rhizoctonia limb rot, cylindrocladium black rot, aspergillus crown rot, and
peanut root-knot nematode [3]. Among these diseases, early leaf spot and late leaf spot are the most typical
peanut diseases due to the warm and humid climate of the country [4]. Leaf spot disease results in increased
defoliation and yield losses of up to 50% [5]. Reduction of quality and quantity of agricultural products like
peanut can possibly be solved through early detection and regular monitoring of diseases. Yet, this is difficult
to be achieved, since detection of peanut diseases such as leaf spot disease is usually done by manually

 ISSN: 2088-8708
Int J Elec & Comp Eng, Vol. 12, No. 3, June 2022: 3005-3012
3006
examining and monitoring some changes in morphology or physical traits. Efficiency in doing monitoring is
a major problem in most areas in the country due to limited manpower to do the task [6].
Meanwhile, the issue of early detection and surveillance is commonly treated using deep
convolutional neural network in plant disease identification because of its nature [7]. The primary purpose of
a deep convolutional neural network or deep convolutional neural network is to learn data characteristics
from convolution operations [8]. The complexity of the architecture and its capability to provide a higher
accuracy model make the deep convolutional neural network, the newest and hottest topic in image
recognition [9], particularly in disease identification [10].
Deep learning convolutional neural networks differ from conventional neural networks (CNN) in the
sense that they have more nodes and more complex means of layer interconnection [11]. Due to its structure,
training a deep convolutional neural network requires high computing power [12] and a large amount of data
in order to attain better results [13]. This technology, however, is not yet used, explored, and examined in the
case of peanut farming [14], particularly in the identification of peanut leaf spot disease due to aspects such
as the availability of sufficient data needed to train and test the model [11].
The purpose of the study is to explore the use of transfer learning algorithm [15], isolation or
background elimination [7], and deep learning optimizers in order to address the insufficient data problem in
creating a model that can identify leaf spot disease [16]. Specifically, it aims to: gather images of peanut or
groundnut leaves, perform pre-processing of the images, perform transfer learning training on the images
using different architectures, optimizers, and learning rate to design classifications, and evaluate the models.
The contributions of this study are the locally collected dataset and the method used in order to achieve the
best model for peanut leaf spot disease identification. This study also explores the use of comprehensive
evaluation on the different pre-trained deep CNN architectures, including the application of the different deep
learning optimizers. The articles are presented as follows: section 2 explains the overall methodology applied
in the study, the experimental set-up, data gathering, data pre-processing, training, and evaluation; section 3
presents the results of the study during training and evaluation; and section 4 provides the conclusions and
recommendations.
2. THEORETICAL BACKGROUND AND RELATED RESEARCHES
2.1. Peanut industry of the Philippines
In the Philippines, the National Capital Region is home to the majority of the country's peanut
industries, while provincial areas in Luzon and Mindanao are home to the remaining micro-scale producers
[1]. Despite of rising local and global demand for peanut finished products, local farmers are still unable to
meet the industry's demands, forcing local peanut product makers to import raw peanuts from other
countries. Inability of the local farmers to meet the demand is attributed to low yielding variety, lack of
high-yielding adaptive cultivars, pests and diseases, poor agronomic practices, climate change, and limited
use of inputs [2]. When it comes to diseases, early and late leaf spot is attributed to defoliation leading to
yield loss of about 50% and more [3]. This disease is common the country due to its humid weather in most
parts of the country in specific month of the year [4].
2.2. Computer vision and agriculture
As computational systems developed, application of machine learning to computer vision achieved
exponential growth leading to the development of novel methodologies and models [17], which now form a
new category, that of deep learning [18]. These methodologies and models are now used for detection and
accurate identification of diseases in grain crops which have great importance for their effective management
in order to guarantee productive and sustainable agriculture [6]. The diagnosis of plant diseases is usually
performed visually and may present flaws due to its laborious and subjective nature [7]. The study of
Barbedo [19] suggests a methods of computer vision with artificial intelligence to automate the process of
detection of diseases in plants. The automatic detection of diseases from images includes, among other
factors, the determination of the most discriminative characteristics for the efficient recognition of the
disease. The use of computer vision particularly deep learning in the field of agriculture only began to take
place in the last couple of years, and to a rather limited extent. It was being utilized to identify diseases in
rice [20], banana [21], avocado [22], grapes [23], coffee [24], abaca [7], and cassava [12].
2.3. Deep convolutional neural network and transfer learning
The accuracy of the network used to detect diseases relies mainly on the complexity and structure of
the architecture [11]. Deep learning convolutional neural network has different architectures, each has
different implementation of the idea of deep convolutional neural network [18]. One of the examples of deep
convolutional network is Visual Geometry Group network architecture which is distinguished by its

Int J Elec & Comp Eng ISSN: 2088-8708 
Peanut leaf spot disease identification using pre-trained deep convolutional … (Urbano B. Patayon)
3007
simplicity. It uses 3×3 convolutional layers stacked on top of one another in increasing depth with max
pooling to reduce volume size. It has a deep of 16 and 19 weight layers [25]. Meanwhile, deep residual
learning framework (ResNet) was developed to address degradation problem due to deeper networks. Instead
of hoping each few stacked layers directly fit a desired underlying mapping, let these layers fit using a
residual mapping [26]. Another solution to degradation problem due to deeper network is addressed by
DenseNet architecture. This architecture simplifies the connectivity pattern and ensures the maximum
information flow [27]. MobileNet, which is also a deep learning convolutional network architecture, is better
suited to mobile and embedded vision applications with limited computational resources. When compared to
a network with conventional convolutions of the same depth in the networks, this design employs depth wise
separable convolutions, which greatly decreases the number of parameters, resulting in a light weight deep
neural network [28]. Furthermore, Inception-v3 is a convolutional neural network architecture designed
based on the Inception group that transports label information lower down the network using label
smoothing, factorized convolutions, and an auxiliary classifier [29]. Xception architecture, on the other hand,
is a version of inception architecture that uses depth wise separable convolutions to replace Inception
modules [30].
Structure of the architectures is one of the main concerns because training a deep convolutional
neural network is difficult to realize since it requires high computing power [12] and a large amount of data
in order to attain better results [13]. The study of Jiang et al. [13], Ramcharan et al. [12], and Sagar and
Dheeba [15] explores the use of transfer learning and deep learning optimizers in order to address the
insufficient data problem and minimal time of training the model.
3. RESEARCH METHOD
3.1. Experimental setup
In conducting the experiment, capturing healthy and leaf spot infected peanut leaves in the sampling
sites using a mobile camera was done first. The background of the image was also considered as part of the
experiment. Images were cleaned, pre-processed, and augmented to eliminate duplication, improve the
quality of the images, and introduce minimal distortion to the images which aids in reducing overfitting at the
training stage. Cleaned images were considered as a dataset and were subjected to training.
During training, the dataset was divided into two groups: healthy and infected peanut leaves.
Weights from ImageNet trained large dataset using VGG16, VGG 19, InceptionV3, MobileNet, DenseNet-
169, Xception, InceptionResNetV2, and ResNet50 architectures were used for retraining. During retraining,
deep learning optimizers such as stochastic gradient descent (SGD) with momentum, adaptive moment
estimation (Adam), root mean square propagation (RMSProp), and adaptive gradient algorithm (Adagrad)
trained on different learning rate were explored. To give an unbiased evaluation of a model retrained on the
training dataset, the candidate model was used to predict the responses in the validation dataset. Using the
test dataset, candidate retrained models were evaluated in order to assess their performance. Each candidate
model was assessed based on its ability to distinguish between healthy and infected peanut leaves in photos.
Figure 1 is the experimental set-up used in the study.
Figure 1. Experiment set-up
3.2. Data gathering
The sampling site of the study is in Liloy, Zamboanga del Norte. The camera used was RealMi 6i
mobile phone with 48-megapixel back camera and 16-megapixel front camera. Upon the taking of images,

 ISSN: 2088-8708
3008
camera was set to manual mode and positioned 0.25 meter above the leaves. In this study, isolation or
background elimination was also considered [13]. Sample images of healthy leaves without background and
with background is shown in Figures 2(a) to 2(d). A total of 1,000 images captured from the sampling site
were validated by experts in plant disease.
(a) (b) (c) (d)
Figure 2. Sample images of (a) healthy leaves without background, (b) with background elimination, (c) leaf
spot infected leaves without background, and (d) with background elimination
3.3. Data pre-processing
Before training, images were labeled based on two groups: healthy and infected leaves. Anchoring
to the study of Mohanty et al. [10], the datasets were divided as follows: 600 images (60%) used for training,
100 images (10%) used for validation, and 100 images (10%) were used for testing. Table 1 is the
distribution of actual samples based on different factors. Before feeding the images into the network, they
were resized to 255×255 in order to reduce the training time [7]. Augmentation such as affine transformation
was also applied in order to increase the dataset and apply slight distortion to the images to reduce over-
fitting [6]. In this study, rescaling and pixel normalization were also applied.
Table 1. Distribution of actual samples based on different factors
Part Type Total Number of Images Training 60% Validation 20% Test 20%
Leaves
Healthy
With Background 250 images
500 images
300
images
50
images
50
images
Without Background 250 images
Infected
With Background 250 images
500 images
300
images
50
images
50
images
Without Background 250 images
3.4. Training of data using transfer learning approach
Transfer learning is based on the idea of reusing a previously trained model on a similar domain
rather than retraining a new model from scratch to improve its performance. In this study, pre-trained
VGG16, VGG 19, InceptionV3, MobileNet, DenseNet-169, Xception, InceptionResNetV2, and ResNet50
were used for training. The architecture is depicted in Figure 3 and begins with several data augmentation
approaches. The images used for training and testing are converted to a 1D array and fed to the dense layer
using the flatten layer. A dropout layer with a dropout rate of 0.7 and a sigmoid activation function was
added for input classification. Adopting the study of Saleem et al. [16], this study used different deep
learning optimizers such as stochastic gradient descent (SGD) with Momentum (beta=0.9) [31], adaptive
moment estimation (Adam), Adagrad, and root mean square propagation (RMSProp) at different learning
rates [16].
Figure 3. The transfer deep learning approach used in the study

3009
3.5. Hardware and software
To train and test the model, python programming language was employed and coded using Jupyter
Notebook. Python libraries such as pandas, NumPy, matplotlib.pyplot, math, sklearn, and Keras built on top
of the tensor flow architecture [7] were utilized. Table 2 is the specification of the hardware used in the
study.
Table 2. Hardware specification used in the study
Parts Specification
Processor Intel(R) Core(TM) i7-9700 CPU @ 3.00GHz
Memory (RAM) 8 gigabytes (GB)
Storage 1 Terabyte (TB)
Graphics NVIDIA GeForce GTX 1080 Ti graphics card
3.6. Evaluation
The typical method for evaluating the performance of artificial neural networks is to divide data into
three sets: training, validation, and test, and then train a neural network on the training set and use the test set
for prediction. Because the testing set's actual outcomes and the model's projected outcomes are both known,
the accuracy of the predictions can be measured. During the testing process, a confusion matrix was used,
and the different models were evaluated on the basis of accuracy and precision, where the formula is shown
in (1) and (2), respectively. Accuracy is defined as how often the classifier is correct, while precision is the
proportions of positive and negative results that are true positive and true negative results. It answers the idea
that when the classifier predicts yes, how often is it correct.
𝐴𝑐𝑐𝑢𝑟𝑎𝑐𝑦 =
𝑇𝑃+ 𝑇𝑁
𝑇𝑃+ 𝑇𝑁+𝐹𝑃+𝐹𝑁
(1)
𝑃𝑟𝑒𝑐𝑖𝑠𝑖𝑜𝑛 =
𝑇𝑃
𝑇𝑃+𝐹𝑃
(2)
Average or mean from complete training was considered in this study [7].
4. RESULTS AND DISCUSSION
The goal of this study is to create models that will identify the presence of disease and healthy
leaves. Explore the use of transfer learning algorithm, isolation or background elimination, and deep learning
optimizers considering the limited data. The succeeding subsections elaborate the results during the training
and evaluation process.
4.1. Training
In this research, weights trained using 10,000+ images from ImageNet were obtained and utilized.
Prior model training to obtain data visualization was performed to assess the performance of the models
trained using different deep learning optimizers. The assessment was done since this research only utilized
small and utterly different dataset from the dataset being used by the pre-trained model.
In this study, the researcher decided to train the model over 100 epochs. The reason to train each
model over 100 epochs is for the researcher to babysit the training process and to identify which stage of the
training process the accuracy and loss start to saddle and to what extent. A saddle point is when the gradient
reaches a plateau, and the training loss becomes harder to improve.
4.2. Evaluation
During testing, the accuracy and precision of the various architectures trained using different deep
learning optimizers were evaluated. The results were tallied and the average or mean of the 100 epochs
training was taken into account. The accuracy and precision are shown in section 4.2.1 and 4.2.2.
4.2.1. Accuracy
The performance of the different pre-trained deep CNN architectures when trained using different
deep learning optimizers in terms of accuracy is illustrated in Figure 4. The research revealed that 6 out of 8
architectures used in the study achieved an accuracy of 90% and above. On the other hand, SGD with
momentum, Adam and RMSProp deep learning optimizers achieved the top 1, top 2, and top 3 in terms of
accuracy in most of the pre-trained deep CNN architectures. In general, DenseNet-169 trained using SGD

 ISSN: 2088-8708
3010
with momentum, Adam, and RMSProp attained the highest accuracy among pre-trained deep CNN
architectures used in the study.
Figure 4. The accuracy of several deep CNN architectures when trained with different deep learning
optimizers
4.2.2. Precision
The performance of the different pre-trained deep CNN architectures when trained using different
deep learning optimizers in terms of precision is illustrated in Figure 5. The research revealed that 6 of the 8
architectures used in the study had precision of 90% or higher. Meanwhile, SGD with momentum, Adam and
RMSProp deep learning optimizers achieved the top 1, top 2, and top 3 in terms of precision in most of the
pre-trained deep CNN architectures. In general, DenseNet-169 trained using RMSProp attained the highest
accuracy among pre-trained deep CNN architectures used in the study.
Figure 5. The precision of various architectures trained using various deep learning optimizers
5. CONCLUSION AND RECOMMENDATION
One of the factors associated with the low production of peanut or groundnut is a disease,
particularly the early leaf spot and late leaf spot which are common in the Philippines due to warm and
humid climate. Studies suggest that early detection and appropriate monitoring are the ways that would
possibly prevent and control the disease. In computer vision deep learning, the deep convolutional neural
network is becoming the preferred method in disease identification and classification due to its impressive
performance. The study explores the use of transfer learning algorithm, isolation or background elimination,
and deep learning optimizers in order to address the insufficient data problem in creating a model that can
identify leaf spot disease.

3011
Using pre-trained VGG16, VGG19, InceptionV3, MobileNet, DenseNet-169, Xception,
InceptionResNetV2, and ResNet50 architectures retrained using deep learning optimizers, the result shows
that DenseNet-169 trained using SGD with momentum, Adam, and RMSProp attained the highest accuracy.
In contrast, DenseNet-169 trained using RMSProp achieved the highest precision among pre-trained deep
CNN architectures as used in the study.
Given the findings, the use of a pre-trained deep convolutional neural network or transfer learning
algorithm, pre-processing techniques, and deep learning optimizers can alleviate problem in data
insufficiency and attain better accuracy and precession results. In addition, this study could be beneficial in
agricultural automation, particularly in creating robotic systems and disease identification systems that can
identify or classify healthy and infected peanut or groundnut plants. Further, it is also recommended to
increase not only the number of training images, but also the number of classes of the different commercially
important Philippine crops in particular. Follow up study will also be recommended to further evaluate the
models when it will be implemented in terms of effectiveness and price in compare with other classification
algorithms.
ACKNOWLEDGEMENTS
This study is supported by the Research Office and conducted in the Machine Learning Laboratory
of Jose Rizal Memorial State University–Tampilisan Campus. Special thanks to Prof. Yhebron J. Lagud,
Prof. Chorla Q. Napigkit, Sir Kerwin Fabillaran, Prof. Jovito Anito, Ryan Caballero, Julakbar Esnain and
Ivan James Gardose.
REFERENCES
[1] J. A. Bernabe and C. C. Sugui, “Promising peanut lines in Northeastern Philippines,” Journal of Food Science and Engineering,
vol. 9, no. 5, pp. 347–358, May 2019, doi: 10.17265/2159-5828/2019.05.005.
[2] H. Saese, G. Fahey, and J. Bafui, “Evaluation of peanut (arachis hypogaea L.) varieties in the lower markham valley of Papua new
guinea,” Agrobacterium Protocols, pp. 347–358, 2006.
[3] L. Guthrie and A. Huber, “Peanut disease identification,” The Peanut Grower. 2015. https://peanutgrower.com/feature/peanut-
disease-identification/ (accessed: Feb. 13, 2021).
[4] B. Shew, “Peanut leaf spots,” NC State Extension Publications. https://content.ces.ncsu.edu/early-leaf-spot-of-peanut-1
(Accessed: Feb. 13, 2021).
[5] G. K. Kishore, S. Pande, and A. R. Podile, “Biological control of late leaf spot of peanut (arachis hypogaea) with chitinolytic
bacteria,” Phytopathology, vol. 95, no. 10, pp. 1157–1165, Oct. 2005, doi: 10.1094/PHYTO-95-1157.
[6] D. I. Patrício and R. Rieder, “Computer vision and artificial intelligence in precision agriculture for grain crops: A systematic
review,” Computers and Electronics in Agriculture, vol. 153, pp. 69–81, Oct. 2018, doi: 10.1016/j.compag.2018.08.001.
[7] U. B. Patayon and R. V. Crisostomo, “Automatic identification of abaca bunchy top disease using deep learning models,”
Procedia Computer Science, vol. 179, pp. 321–329, 2021, doi: 10.1016/j.procs.2021.01.012.
[8] F. Albu, A. Mateescu, and N. Dumitriu, “Architecture selection for a multilayer feedforward network,” International Conference
on Microelectronics and Computer Science, pp. 131–134, 1997, doi: 10.1.1.703.2957.
[9] R. Gandhi, S. Nimbalkar, N. Yelamanchili, and S. Ponkshe, “Plant disease detection using CNNs and GANs as an augmentative
approach,” in 2018 IEEE International Conference on Innovative Research and Development (ICIRD), May 2018, pp. 1–5, doi:
10.1109/ICIRD.2018.8376321.
[10] S. P. Mohanty, D. P. Hughes, and M. Salathé, “Using deep learning for image-based plant disease detection,” Frontiers in Plant
Science, vol. 7, pp. 1–10, Sep. 2016, doi: 10.3389/fpls.2016.01419.
[11] M. P. Vaishnnave, K. Suganya Devi, and P. Ganeshkumar, “Automatic method for classification of groundnut diseases using deep
convolutional neural network,” Soft Computing, vol. 24, no. 21, pp. 16347–16360, Nov. 2020, doi: 10.1007/s00500-020-04946-0.
[12] A. Ramcharan, K. Baranowski, P. McCloskey, B. Ahmed, J. Legg, and D. P. Hughes, “Deep learning for image- based cassava
disease detection,” Frontiers in Plant Science, vol. 8, Oct. 2017, doi: 10.3389/fpls.2017.01852.
[13] H. Jiang, Z. P. Xue, and Y. Guo, “Research on plant leaf disease identification based on transfer learning algorithm,” Journal of
Physics: Conference Series, vol. 1576, no. 1, Jun. 2020, doi: 10.1088/1742-6596/1576/1/012023.
[14] M. Dong, S. Mu, T. Su, and W. Sun, “Image recognition of peanut leaf diseases based on capsule networks,” Artificial
Intelligence, pp. 43–52, 2019, doi: 10.1007/978-981-32-9298-7_4.
[15] A. Sagar and D. Jacob, “On using transfer learning for plant disease detection,” bioRXiv, 2020, doi:
10.13140/RG.2.2.12224.15360/1.
[16] M. H. Saleem, S. Khanchi, J. Potgieter, and K. M. Arif, “Image-based plant disease identification by deep learning meta-
architectures,” Plants, vol. 9, no. 11, p. 1451, Oct. 2020, doi: 10.3390/plants9111451.
[17] Y. LeCun, “LeNet-5, convolutional neural networks,” MNIST demos on Yann LeCun’s website. http://yann.lecun.com/exdb/lenet
(accessed: Aug. 24, 2021).
[18] S. J. Russell and P. Norvig, Artificial intelligence: a modern approach. Hoboken: Pearson, 2021.
[19] J. G. A. Barbedo, “A review on the main challenges in automatic plant disease identification based on visible range images,”
Biosystems Engineering, vol. 144, pp. 52–60, Apr. 2016, doi: 10.1016/j.biosystemseng.2016.01.017.
[20] R. R. and D. Park, “A multiclass deep convolutional neural network classifier for detection of common rice plant anomalies,”
International Journal of Advanced Computer Science and Applications, vol. 9, no. 1, 2018, doi: 10.14569/IJACSA.2018.090109.
[21] H. P. M. Lim and M. R. J. E. Estuar, “Microscopic fusarium detection and verification with convolutional neural networks,” in
Proceedings of the 2018 International Conference on Machine Learning Technologies - ICMLT ’18, 2018, pp. 48–52, doi:
10.1145/3231884.3231892.
[22] A. Salgadoe, A. Robson, D. Lamb, E. Dann, and C. Searle, “Quantifying the severity of phytophthora root rot disease in avocado
trees using image analysis,” Remote Sensing, vol. 10, no. 2, p. 226, Feb. 2018, doi: 10.3390/rs10020226.

 ISSN: 2088-8708
3012
[23] A. Cruz et al., “Detection of grapevine yellows symptoms in Vitis vinifera L. with artificial intelligence,” Computers and
Electronics in Agriculture, vol. 157, pp. 63–76, Feb. 2019, doi: 10.1016/j.compag.2018.12.028.
[24] E. Hitimana and O. Gwun, “Automatic estimation of live coffee leaf infection based on image processing techniques,” in
Computer Science & Information Technology ( CS & IT ), Feb. 2014, pp. 255–266, doi: 10.5121/csit.2014.4221.
[25] A. Rosebrock, “ImageNet: Vggnet, ResNet, inception, and xception with keras,” PyImageSearch. 2021,
https://www.pyimagesearch.com/2017/03/20/imagenet-vggnet-resnet-inception-xception-keras/ (accessed: Aug. 24, 2021).
[26] K. He, X. Zhang, S. Ren, and J. Sun, “Deep residual learning for image recognition,” in 2016 IEEE Conference on Computer
Vision and Pattern Recognition (CVPR), Jun. 2016, pp. 770–778, doi: 10.1109/CVPR.2016.90.
[27] G. Huang, Z. Liu, L. Van Der Maaten, and K. Q. Weinberger, “Densely connected convolutional networks,” in 2017 IEEE
Conference on Computer Vision and Pattern Recognition (CVPR), Jul. 2017, pp. 2261–2269, doi: 10.1109/CVPR.2017.243.
[28] Y. Harjoseputro, I. P. Yuda, and K. P. Danukusumo, “MobileNets: efficient convolutional neural network for identification of
protected birds,” International Journal on Advanced Science, Engineering and Information Technology, vol. 10, no. 6, Dec. 2020,
doi: 10.18517/ijaseit.10.6.10948.
[29] C. Szegedy, V. Vanhoucke, S. Ioffe, J. Shlens, and Z. Wojna, “Rethinking the inception architecture for computer vision,” in
2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Jun. 2016, pp. 2818–2826, doi:
10.1109/CVPR.2016.308.
[30] F. Chollet, “Xception: deep learning with depthwise separable convolutions,” in 2017 IEEE Conference on Computer Vision and
Pattern Recognition (CVPR), Jul. 2017, pp. 1800–1807, doi: 10.1109/CVPR.2017.195.
[31] A. Gulli and S. Pal, Deep learning with Keras: implement neural networks with Keras on Theano and TensorFlow. Birmingham,
UK: Packt Publishing, 2017.
BIOGRAPHIES OF AUTHORS
Urbano B. Patayon is an Instructor I at Jose Rizal Memorial State University
(JRMSU). He is the program head of Computer Science Department of Jose Rizal Memorial
State University-Tampilisan Campus. He obtained his Bachelor of Science in Computer
Science degree at Jose Rizal Memorial State University as Academic Scholar and finished his
Master of Science in Computer Science at Mindanao State University-Iligan Institute of
Technology as CHED K-12 scholar. He is an educator and a researcher at heart. He has
presented researches in national and international fora and had published scientific papers in
SCOPUS indexed journals. His research interest focus in artificial intelligence, learning
science, internet of things, data science, and software engineering. He can be contacted at
email: urbanopatayon@jrmsu.edu.ph.
Renato V. Crisostomo is an Assistant Professor and a member of the Institute
Curriculum Committee for Computer Science Department of Mindanao State University-
Iligan Institute of Technology. His research interest focuses in artificial intelligence,
programming languages, natural language processing, and information security. He can be
contacted at email: renato.crisostomo@g.msuiit.edu.ph.

Peanut leaf spot disease identification using pre-trained deep convolutional neural network

More Related Content

Similar to Peanut leaf spot disease identification using pre-trained deep convolutional neural network (20)

More from IJECEIAES (20)

Recently uploaded (20)

Peanut leaf spot disease identification using pre-trained deep convolutional neural network