University of Gondar

Faculty of Computing
Department of Information Technology
Research proposal
Detection and Grading of Maize Leaf Diseases Using Deep Convolutional
Neural Networks

Prepare By:

Dagne Walle
Submitted To: Department of IT

NOV 20, 2020

Table of Contents
1. Introduction .......................................................................................................................................... 1
1.1. Background ................................................................................................................................... 2
1.2. Statement of Problem ................................................................................................................... 3
1.3. Objectives of the Study ................................................................................................................. 4
1.4. Scope and limitation of study ................................................................................................. 5
1.5. Research Methodology ..................................................................................................... 6
1.5.1 . Litrature review ......................................................................................................... 6
1.5.2 Dataset preparation ................................................................................................ 6
1.5.3 Feature Learning ............................................................................................................... 7
1.5.4 Implementation procedure and tools ............................................................................................ 8
1.5.5 Performance Evaluation .............................................................................................. 9
1.6 Contribution of the Study ............................................................................ 9
1.7 Work breakdown and budget............................................................................................. 10
1.8 Reference....................................................................................................................................... 11

i
1. Introduction
1.1. Background

Maize is one of the most important cereal crops used in the human diet in large parts of the world
and it is an important food component for livestock. It is not only an important for human, but
also a basic food for animal feeds. It is an important food grain crop in Ethiopia and is produced
throughout the country under diverse environments [1]. In Ethiopia, maize is one of the principal
cereal crops ranking first in total production and productivity. Maize is everything for the
Ethiopian maize farmers [3]. There are many Biotic disease that infect a maize production in
Ethiopia and the following are common and emerging.

Common leaf rust (CLR), caused by Puccinia sorghi Schwein, fungus is another important
disease of maize in Ethiopia that is widely distributed throughout the major maize growing
regions of the country. However, the importance varies from place to place. It is more severe in
the southern mid-altitude and the highland sub-humid maize growing agro-ecologies of the
country. There is a pustule on the surface of the leaf on both side of the leaf and this disease
emerged at the maturity stage of the maize. Yield loss as high as 23% was reported due to
common maize rust on susceptible cultivars under favorable field conditions in Ethiopia. Grey
leaf spot was first reported in Ethiopia in 1997 in the border of west Wellega and Ilubabor zones,
of western Ethiopia and increased prevalence of grey leaf spot in the major maize producing
regions of Western, Southern and Northwestern parts of Ethiopia [11]. This disease is caused by
fungus Cercospora zeae-maydis. The frequent warm rainy periods create favorable conditions for
this fungal disease. Gray leaf spot initially appears as spots on the lower of leaves that are round,
tan in color, and have a dark brown border. When the leaves are humidity is high, the leaf spots
turn gray and fuzzy. The disease is significant since it rapidly destroys foliage when the maize is
near at stage of grain maturity [11]. Later, the disease was spread to all major maize producing
mid altitude subhumid agroecology of the country. To date, the disease is one of the most
important threats to maize production in the country, causing yield losses as high as 29.1% [12].

Turcicum leaf blight (TLB) caused by Exserohilum turcicum (Pass.) is known to infect maize
from the seedling to maturity stage. Leaf blight disease affected by fungus. The initial symptoms
are small elliptical spots on the leaves as grayish green with water-soaked lesions parallel to leaf
margins. is a major disease affecting maize production in western Ethiopia. The disease can
cause loss of the nutritive value of maize. TLB is one of the major maize diseases having wide
distribution and high economic importance in Ethiopia and spread by wind and rain. The disease
is prevalent from low land humid through highland humid agr o-ecologies during the wet rainy
seasons.
1
occurs sporadically in most temperate, humid areas where maize is grown. In Ethiopia, maize
leaf blight recorded from 500 to 2650 m above sea level [3] . Yield losses as high as 70% have
been attributed to Turcicum leaf blight in Ethiopia [10]. Maize lethal necrosis (MLN) disease has
emerged as a great threat of maize production in Ethiopia and caused by virus and infect from
seeding to maturity stage or it make the maize leaf dry at all. Leaves of infected plants become
yellow from the tip and margins to the center. Ears and leaves dry up and sometimes look like a
mature plant. The whole plant dies and maize cobs remain without kernels and in the early
stages, MLND causes long yellow stripes on leaves. In 2014 and 2015, surveys conducted in
Oromia, SNNP, Benishangul Gumuz, Amahara and Tigray regions showed that the two viruses
were present in a significant number of plant and The report of maize lethal necrosis disease in
Ethiopia is of high level concern for producers in the country because of the potential of an
epidemic that can devastate maize seed and grain production and According to that survey
report MLND severity in SNNP was the highest with a mean rate of 3.5, followed by Oromia,
Benishangul Gumuz, Amahara and Tigray with mean severities 3.2, 2.5, 2, and 1.5, respectively
and It is estimated that severely affected areas can experience a massive yield loss of over 95%
and this will affect total maize yield produced in the country[23] . Maize streak disease is caused
by Maize streak virus (MSV). Similar epidemics has occurred in south western Ethiopia in 2013,
which caused yield losses of 50 to 100% on farmers‟ fields (Ethiopian Institute of Agricultural
Research). In Ethiopia, this disease was formerly known only in Gambella (the western low land
sub-humid plains bordering South Sudan). But in recent years, the disease is becoming very
important in the mid-altitude agroecology of Ethiopia and posing a significant threat to maize
production in the country and Most commercial varieties currently under production are found to
be susceptible to the virus [24]. As the disease progresses, mild to severe chlorotic streaks
develop along leaf veins and stunting occurs. Maize plants infected at an early stage produce
undersized, misshaped cobs or give no yield at all. Brown spot maize leaf disease caused by
fungus Physoderma maydis of maize has been observed to cause considerable crop damage in
warm humid areas of western and southern regions of Ethiopia at Gofa (Saula), Shallo and
Hawassa [28]. Its most observable damage is the numerous big brown spots on the leaves which
can kill the whole leaf. When infection occurs in the seed, unfilled grains or spotted or
discolored seeds are formed and the disease occurred at the matured stage of maize

2
 1.2. Statement of the Problem

Maize leaf disease identification is one of the most basic and important activities in agriculture.
In most cases, identification is performed manually, either visually or by microscopy. The
problem with visual assessment is that, being a subjective task, it is prone to psychological and
cognitive phenomena that may lead to bias, optical illusions and, ultimately, to error. On the
other hand, laboratorial analyses such as molecular, immunological or pathogen culturing-based
approaches are often time consuming, failing to provide answers in a timely manner. In this
context, it is compelling to develop automatic methods capable of identifying maize leaf diseases
in a rapid and reliable way[2]. The vast majority of automatic methods proposed so far rely on
digital images, which allows the use of very fast techniques. However, intrinsic and extrinsic
factors mean these methods remain too error prone, which is the motivation for developing a
model that detecting and grading a maize leaf disease to identified maize leaf disease. The naked
eye observation of experts is the main approach adopted in practice for detection and
identification of plant diseases [19].

However, this requires continuous monitoring of experts which might be prohibitively expensive
in large farms. Further, in some developing countries, farmers may have to go long distances to
contact experts, this makes consulting experts too expensive and time consuming [3].Automatic
detection of maize diseases is an essential research topic as it may prove benefits in monitoring
large fields of crops, and thus automatically detect the symptoms of diseases as soon as they
appear on plant leaves . Therefore; looking for fast, automatic, less expensive and accurate
method to detect plant disease cases is of great realistic significance to increase maize product.
Imaging and machine learning based recognition and grading of plant or cereals diseases can
provide evidences to identify and treat the diseases in its early stages [8].

Maize leaf diseases diagnosis is very essential in earlier stage in order to cure and control them.
There are different traditional mechanisms to detect and grade maize leaf diseases with human
operator, chemicals etc. However, the traditional mechanisms to detection maize leaf diseases
have their own draw backs, such as expensiveness, prone to error, takes time, inconsistent,
require expertise in the plant diseases, specialized instruments and inefficient[9].Currently,
infectious diseases reduce the potential yield by an average of 40% with many farmers in the
developing world experiencing yield losses as high as 100%[17].
3
previous researchers are conducted a research on different maize leaf disease and they use a
Visual observation to collect maize leaf image from farm area and for detect and classifying
maize leaf disease with very small dataset .However they are not concidering the most harmful
and emerging Ethiopian maize leaf disease using deep CNN with respecting to thire level of
damage and it is difficult to collect disease images from farm area by previous researchers
without maize pathologist due to the following challenges [2]. The background often contains
elements that can make it very difficult to correctly segment the region of interest where the
symptoms are manifest. Capture conditions are difficult to control, which may cause the images
to present characteristics that are difficult to predict and make the disease identification more
challenging. Most symptoms do not have well defined boundaries, rather gradually fading into
normal tissue, making it difficult to clearly define which are the healthy and diseased regions. A
given disease may possess very different characteristics depending on its stage of development,
and sometimes on where it is located on the maize leaf. Symptoms produced by different
diseases may be present simultaneously, manifesting either physically separated or combined
into a “hybrid” symptom that may be difficult to identify. Symptoms produced by different
diseases may be visually similar, which forces the methods to rely on very tenuous differences to
discriminate among them.
To address the above-mentioned challenges this study will use a dataset prepared by maize
pathologist on plant village public repository dataset . Therefore, this study is, motivated to build
automated system, which is faster and more accurate using deep convolution neural network to
detecting and grading maize leaf disease and to control disease at early stage before damage the
maize crops After the end of this study the following question should be answered

1. Is it possible to detect and grade maize diseases using deep CNN?

2. How to design a deep CNN model that able to detect and grade maize’s leaf disease?
3. How to develop a prototype of the proposed system and increase its performance using
validation datasets?
4. How to test and evaluate the proposed system using test datasets?
4
1.3 Objective of the study

1.3.1 General objective

The main objective of this study is to design and develop automatic detection and grading of
maize leaf diseases using deep convolutional neural networks.
1.3.2 Specific objective
To achieve the general objective of the research the following specific objective should be
considered
 To collecting images of maize leaf disease and grading according to its severity
 To analyzing the current image recognition model developing using deep learning
 To developing a prototype of the proposed system and increasing its performance using
validation datasets
 To test and evaluate the system using test datasets
 To designing a deep convolutional neural network model for detecting and grading
maize’s leaf disease

1.4 Scope and limitation of the study

This study focuses on automatic detection and grading of Ethiopian most harmful maize leaf
diseases using deep CNN. They are maize leaf blight (LB), Common leaf rust (CLR), gray leaf
spot (GLS), maize leaf streak (MLS), maize leaf lethal necrosis (MLLN) and maize brown leaf
spot (BLS) this study has (six leaf disease grade and one healthy) and grading an Ethiopian
common and emerging maize leaf disease into seven total grades according to the disease
severity. Because those disease distributed across the entire region of Ethiopia and have high
level of damage on maize agriculture and loose farmers maize yield and this study not include
non-common and non-emerging maize leaf disease. however there are reported more than 72
maize leaf disease found in Ethiopia[3]. All maize disease are not including due to dataset
problems but most common and emerging disease that occurred in Ethiopian maize farming are
included.

5
 1.5 Research methodology

This study will use different procedures, techniques, tools, and documentation aids which will
help to undertake and implement the research work. This research is initiated to build a model
for Maize leaf diseases detection and grading using deep convolution neural network to early
control the stage of the maize disease. under this study, literature review, dataset preparation,
implementation procedures, tools and performance evaluation metrics are will be used to
accomplish the research work and this study is design science research.

1.5.1 Literature Review

To effectively completion of this research several worldwide and nearby research work will

have review from the starting of the study until its completion. journal articles, conference
papers, reports, books and  will have review for doing the research to achieve the  objective of
the study and to reply the research  questions mentioned above and it use a deep convolution
neutral network algorithm that used to detecting and grading maize leaf infections to control
the disease early and to improve maize product for maize farmers].

1.5.2 Dataset preparation

This study uses a Plant Village Dataset expertly prepared and organized by Penn State
university Plant pathology. The datasets are available on https://www.kaggle.com and
www.plantvillage.org websites .The image used for research and development that empowers
smallholder farmers and seeks to lift them out of poverty using cheap, affordable technology to
reduce yield losses to crops diseases. Maize leaf disease images for training and testing the
model to detect and grading the maize leaf disease and the data are collected from plant village
and Kaggle repository.

6
The images of maize leaf will be preprocessing to reduce data noise. Plant Village is a web
platform developed at Penn State that also connects the international community of public
funded scientists to farmers around the world and assist helps country level extension workers to
help farmer. Three years ago, co-founded an online platform dedicated to crop health and crop
diseases, called Plant Village and confirmed the identity of diseases by having the expert plant
pathologists determine the disease states and all images into the Plant Village database using the
diagnosis from the experts only expertly identified leaves are present in the database [17].

Release of over 50,000 expertly curated images on healthy and infected leaves of crops plants
through the existing online platform Plant Village and Ethiopian Agricultural Transformation
Agency also a partner of plant village and Plant Village is an open access online knowledge base
developed at Penn State University that make available content on crop health. Kaggle, a
subsidiary of Google, is an online community of data scientists and machine learning physicians.
Kaggle allows users to find and publish datasets, explore and build models in a web-based data-
science environment, work with other data scientists and machine learning engineers, and enter
competitions to solve data science challenges. Therefore, this study uses a dataset available on
plant village ,Kaggle and google website.

1.5.3 Feature learning

In feature learning process we have two component image preprocessing (image augmentation
and segmentation), and deep learning (feature learning and classification). first layer in CNN is a
convolutional layer. The input to the first convolutional layer is an RGB image whereas the
feature map is the input to the other convolutional layer. In order to get a feature map, the input
image is convolved with or by the kernel filter to extract important features or find the most
representative feature in a given region. During convolution, we have considered important
parameters: padding, stride, number of filter And filter size. Learning process contains three
phases: - training, validation and testing phase.

7
 1.5.4 Implementation procedure and tools

Deep convolution neural network (CNN) algorithm will be utilizing to improve the performance
of the model. In recent years, one of the most significant innovations in machine learning is deep
learning. Deep convolutional neural networks (CNN) have become a hot field in plant disease
detection and grading to improve the performance of the model [18].

The main types of neural networks used for multidimensional signal processing are deep
convolutional neural networks (CNNs). With the development of large data age, Convolutional
neural networks (CNNs) with more hidden layers have more complex network structure and
more powerful feature learning and feature expression abilities than traditional machine learning
methods and it can make the training can use a higher learning rate, reduce training time; at the
same time reduce the over-fitting, improve the accuracy rate[25] .

Therefore, this study uses deep convolution neural network algorithm. Python programming is
used for image preprocessing, feature extraction and to build the model. So, we will use
Anaconda environment, which is an exposed source delivery of Python programming language
for scientific computing. We will also install, configure and use Kera’s and TensorFlow in this
environment on windows operating system. Kera’s is a free source CNN library written in
Python. It is easily running on top of TensorFlow. TensorFlow is a representative math library
and is also used for scientific computing tenders such as neural networks and these TensorFlow
as a backend. python code is written in Spyder IDE .

8
 1.5.5 Performance Evaluation

Evaluation of the developing model will be using the test dataset of maize leaf disease image. The
performance of the model is evaluate based on the number of maize leaf disease correctly
detecting and grading and how many of them are detecting and grading incorrectly. This propose
model will be evaluated using confusion matrix and ROC curve metrics. Therefore, this study will
be use evaluation metrices to measure the performance of the model detection and grading of
maize disease according to its severity.

1.6 Contribution of the Study

This study is important specially for maize farmers and farm experts to control the disease at early
stage and to increase the maize grain yield and in order to enable agricultural producers to make
quick and reasonable judgments about crop disease information. The solution for early disease
detection is a deep convolutional neural network, which is now the most robust technique for
maize leaf image grading and detection. Deep convolution neural network algorithm has high
processing speed and high grading and detection accuracy.

Crop-disease detection system provides clear benefit in monitoring of large fields, as this is the
only approach that provides a chance to detect diseases at an early stage. Therefore, diagnosis is
one of the most important aspects of a maize pathologist's training. Without proper identification
of the disease and the disease-causing agent, disease control measures can be a waste of time and
money and can lead to further crop losses. Proper disease diagnosis is therefore vital to growth the
farmer maize grain yield

9
1.7 Work plane
No. Activities Duration of time 2020
Nov Dec Jan Feb Mar Apr May June
1. Literature Review
2. Research proposal
development
3. Research proposal
Defense
4. Preparation of
Experimental setup
5. Configuring experimental
Environment
Data collection from plant
6. village website
Data analysis, processing
7. image and
Train, test a model
8. Draft Report Submission
9. Final Report Submission

1.8 Budget breakdown

No. Description Estimated Cost (ETB)
1. Hard Disk, printing and stationary 4000
Materials
2. Transport and other basic need at that time 9000
of contact with advisor
To buy computer to processing the data and build
3. the model 110000
Total 24000

10
Reference
[1] A.Kidest “Growth, productivity, and nitrogen use efficiency of maize,” 2013.
[2] J. Garcia and A. Barbedo, “ScienceDirect A review on the main challenges in automatic plant
disease identification based on visible range images,” Biosyst. Eng., vol. 144, pp. 52–60, 2016, doi:
10.1016/j.biosystemseng.2016.01.017.
[3] E.Alehegn, “maize leaf diseases recognition and classification based on imaging and machine
learning techniques,” 2017.
[4] Gero, J. S. (1999). ‘Research Methods for Design Science Research: Computational and Cognitive
Approaches. In Proceedings of ANZAScA. Australia.’
[5] M. A. Austin, “Engineering Programming in MATLAB : A Primer,” 2000.
[6] K. T. et al., “Major biotic maize production stresses in Ethiopia and their management through host
resistance,” African J. Agric. Res., vol. 13, no. 21, pp. 1042–1052, 2018, doi: 10.5897/ajar2018.13163.
[8] M. Reyalat, M. Braik, and Z. Alrahamneh, “Fast and Accurate Detection and Classification of Plant
Diseases Fast and Accurate Detection and Classification of Plant Diseases,” no. March 2015, 2011, doi:
10.5120/2183-2754.
[9] “Agricultural plant Leaf Disease Detection Using Image Processing,” vol. 2, no. 1, pp. 599–602,
2013.
[10] Gowda KP, Shetty H, Gowda B, Prakash H, Lal S (1992). “Comparison of two methods for
assessment of yield loss due to turcicum leaf blight of maize. Indian Phytopathology” 45(3):316-320.
[11] A. Nega and F. Lemessa, “Distribution and Importance of Maize Grey Leaf Spot Cercospora zeaemaydis
(Tehon and Daniels) in South and Southwest Ethiopia,” J. Plant Pathol. Microbiol., vol. 7, no. 7, 2016, doi:
10.4172/2157-7471.1000362..
[12] Wegary D, Kitaw D, Demissie G (2004). ‘Assessment of losses in yield and yield components of
maize varieties due to grey leaf spot. Pest Management Journal of Ethiopia’ 8:59-
[13] Tilahun T, Wagary D, Demissie G, Negash M, Admassu S, Jifar H (2012). Maize Pathology
Research in Ethiopia in the 2000s: A Review. ‘Meeting the Challenges of Global Climate Change and
Food Securit through Innovative Maize Research, Addis Ababa’ P. 193.
[15] Shiferaw B, Prasanna BM, Hellin J, Bänziger M (2011). Crops that feed the world 6. Past
successes and future challenges to the role played by maize in global food security. Food Security
3(3):307-327.
[16] CSA (2017). ‘Centeral Statistical Agency Agricultural Sample Survey, Report on area and
production of major crops.’ www.csa.gov.et. Accessed 20 December 2017.
[17] D. P. Hughes and M. Salathe, “An open access repository of images on plant health to enable the
development of mobile disease diagnostics,” 2015.
[18] N. N. Model, C. N. Networks, D. N. Network, C. Layer, H. Wu, and J. Zhao, “Deep Convolutional
Neural Networks 13th International Symposium on Process Systems Engineering ( PSE,” 2018.)
[19]. Weizheng, S., Yachun, W., Zhanliang, C., and Hongda, W. (2008). Grading Method of Leaf Spot
Disease Based on Image Processing. In Proceedings of the 2008 international Conference on Computer
11 Science and Software Engineering - Volume 06 (December 12 - 14,2008). CSSE. IEEE Computer
Society, Washington, DC,491-494. DOI= http://dx.doi.org/10.1109/CSSE. 2008.1649.
[20]. Prasad Babu, M. S. and Srinivasa Rao , B. (2010) L eaves recognition using back-propagation
neural network - advice for pest and disease control on crops.Technical report, Department of Computer
Science &Systems Engineering, Andhra University, India. Downloaded from www.indiakisan.net on May
2010.
[21]. Camargo, A. and Smith, J. S., (2009). An imageprocessing based algorithm to automatically
identify plant disease visual symptoms, Biosystems Engineering, Volume 102, Issue 1, January 2009,
Pages 9-21, ISSN 1537-5110, DOI: 10.1016/j.biosystemseng.2008.09.030.
[22]. Hillnhuetter, C. and A.-K. Mahlein, Early detection and localisation of sugar beet diseases: new
approaches, Gesunde Pfianzen 60 (4) (2008), pp. 143–149.
[23] M. H. Girma Demissie, Temesgen Deressa, “Prevalence, Distribution and Impact of Maize Lethal
Necrosis Disease (MLND) in Ethiopia,” Pest Manag. J. Ethiop., vol. 18, no. March 2016, pp. 37–49,
2016.
[24] K. T. et al., “Major biotic maize production stresses in Ethiopia and their management through host
resistance,” African J. Agric. Res., vol. 13, no. 21, pp. 1042–1052, 2018, doi: 10.5897/ajar2018.13163.
[25] A. A. M. Al-Saffar, H. Tao, and M. A. Talab, “Review of deep convolution neural network in image
classification,” Proceeding - 2017 Int. Conf. Radar, Antenna, Microwave, Electron. Telecommun.
ICRAMET 2017, vol. 2018-Janua, no. March 2020, pp. 26–31, 2017, doi: 10.1109/ICRAMET.2017.8253139.
[26] M. Debela, M. Dejene, and W. Abera, “Management of Turcicum leaf blight [ Exserohilum turcicum
( Pass .) Leonard & Suggs ] of maize ( Zea mays L .) through integration of host resistance and fungicide
at Bako , Western Ethiopia,” vol. 11, no. January, pp. 6–22, 2017, doi: 10.5897/AJPS2016.1449.
[27] A. Ababa, Enhancing the Contribution of Maize to Food Security in Ethiopia Enhancing the
Contribution of Maize to Food Security in Ethiopia, no. November. 2001.
[28] T. Tilahun, D. Wagary, G. Demissie, M. Negash, S. Admassu, and H. Jifar, “Maize
Pathology Research in Ethiopia in the 2000s : A Review Maize Pathology Research in Ethiopia
in the 2000s : A Review,” no. March 2015, 2012.
12