AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf ·...

16
AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF DISEASE USING DEEP CONVOLUTIONAL NEURAL NETWORK Dr.L. Malathi1, P. Yogashree2, A. Thamaraiselvi3 1Associate Professor, Department of Computer Science and Engineering, Vivekanandha College of Engineering for Women, Tiruchengode, Namakkal,Tamilnadu, India 2 P.G.Scholar, Department of Computer Science and Engineering, Vivekanandha College of Engineering for Women, Tiruchengode, Namakkal, Tamilnadu, India 3 Assistant Professor, Department of Computer Science Engineering, Vivekanandha College of Engineering for Women, Tiruchengode, Namakkal, Tamilnadu, India ABSTRACT- Agriculture profitability is something on which economy profoundly depends in India. This is the one reason that infections discovery in plants assumes a significant job in agribusiness field, as having ailment in plants are very normal. On the off chance that appropriate consideration can't right now, it causes genuine impacts on plants and because of which particular item quality, amount or profitability is influenced. Presently a-days profound learning is turning into the standard strategy for picture characterization. Ongoing advancements in Deep Neural Networks have permitted analysts to radically improve the exactness of item location. Effectively scarcely any engineering has been created by the analysts for proficient characterization plant ailment which incorporates Faster Region-based Convolutional Neural Network (Faster R-CNN), Region-based Fully Convolutional network(R-FCN), and Single Shot Multibox Detector (SSD). Alex Net, Google Net and VCC-16 are the surely understands Deep Convolutional Neural Network design for general Image arrangement. In any case, these engineering doesn't perform well, when numerous sicknesses influencing a similar leaf. To conquer this, A Hybrid Deep Convolutional Neural Network design with division is proposed right now which contains five convolutional layer, five pooling layer and two completely associated layer. Profound convolutional Neural Networks are most ordinarily used to break down visual symbolism and are habitually working off camera in picture grouping. Key Words: Deep Learning, Alex Net, Google Net, Plant Disease Prediction, etc. I. INTRODUCTION Yields are influenced by a wide assortment of maladies, particularly in tropical, subtropical, and mild districts of the world. Plant infections include composite associations between the host plant, infection, and its vector. Environmental change fundamentally influences provincial atmosphere factors, for example, dampness, temperature, and precipitation, that subsequently fill in as a vector where pathogens, infection, and sicknesses can obliterate a yield, and in this manner premise the immediate effects on the populace, for example, financial, wellbeing, and vocation impacts. A prior ID of infection is these days a moving methodology and should be treated with extraordinary consideration. The exploration has center around the recognizable proof and acknowledgment of infections that influence tomato, banana, maize, sugarcane, and rice plants. So for the acknowledgment of plant sicknesses the innovation utilized for the exploration is Convolutional Neural Network for the forecast in the plant leaves. Convolutional Neural systems are layout to the procedure of information through various layers of clusters. This sort of neural systems is utilized in applications like picture acknowledgment or face acknowledgment. The most punctual contrast among CNN and some other normal neural system is that CNN accepts contribution as a two-dimensional cluster and control straightforwardly on the pictures as opposed to concentrating on highlight creation which other neural systems center around. CNN or Convolutional Neural Networks use pooling layers, which are the layers, situated following CNN statement. Journal of Xi'an University of Architecture & Technology Volume XII, Issue V, 2020 ISSN No : 1006-7930 Page No: 674

Transcript of AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf ·...

Page 1: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

AUTOMATIC DETECTION AND

CLASSIFICATION OF PLANT LEAF DISEASE

USING DEEP CONVOLUTIONAL NEURAL

NETWORK

Dr.L. Malathi1, P. Yogashree2, A. Thamaraiselvi3

1Associate Professor, Department of Computer Science and Engineering, Vivekanandha College of Engineering

for Women, Tiruchengode, Namakkal,Tamilnadu, India 2P.G.Scholar, Department of Computer Science and Engineering, Vivekanandha College of Engineering for

Women, Tiruchengode, Namakkal, Tamilnadu, India 3Assistant Professor, Department of Computer Science Engineering, Vivekanandha College of Engineering for

Women, Tiruchengode, Namakkal, Tamilnadu, India

ABSTRACT- Agriculture profitability is something on which economy profoundly depends in India. This

is the one reason that infections discovery in plants assumes a significant job in agribusiness field, as

having ailment in plants are very normal. On the off chance that appropriate consideration can't right

now, it causes genuine impacts on plants and because of which particular item quality, amount or

profitability is influenced. Presently a-days profound learning is turning into the standard strategy for

picture characterization. Ongoing advancements in Deep Neural Networks have permitted analysts to

radically improve the exactness of item location. Effectively scarcely any engineering has been created by

the analysts for proficient characterization plant ailment which incorporates Faster Region-based

Convolutional Neural Network (Faster R-CNN), Region-based Fully Convolutional network(R-FCN), and

Single Shot Multibox Detector (SSD). Alex Net, Google Net and VCC-16 are the surely understands Deep

Convolutional Neural Network design for general Image arrangement. In any case, these engineering

doesn't perform well, when numerous sicknesses influencing a similar leaf. To conquer this, A Hybrid

Deep Convolutional Neural Network design with division is proposed right now which contains five

convolutional layer, five pooling layer and two completely associated layer. Profound convolutional

Neural Networks are most ordinarily used to break down visual symbolism and are habitually working

off camera in picture grouping.

Key Words: Deep Learning, Alex Net, Google Net, Plant Disease Prediction, etc.

I. INTRODUCTION

Yields are influenced by a wide assortment of maladies, particularly in tropical, subtropical, and mild

districts of the world. Plant infections include composite associations between the host plant, infection, and its

vector. Environmental change fundamentally influences provincial atmosphere factors, for example, dampness,

temperature, and precipitation, that subsequently fill in as a vector where pathogens, infection, and sicknesses

can obliterate a yield, and in this manner premise the immediate effects on the populace, for example, financial,

wellbeing, and vocation impacts. A prior ID of infection is these days a moving methodology and should be

treated with extraordinary consideration. The exploration has center around the recognizable proof and

acknowledgment of infections that influence tomato, banana, maize, sugarcane, and rice plants. So for the

acknowledgment of plant sicknesses the innovation utilized for the exploration is Convolutional Neural Network

for the forecast in the plant leaves. Convolutional Neural systems are layout to the procedure of information

through various layers of clusters. This sort of neural systems is utilized in applications like picture

acknowledgment or face acknowledgment. The most punctual contrast among CNN and some other normal

neural system is that CNN accepts contribution as a two-dimensional cluster and control straightforwardly on

the pictures as opposed to concentrating on highlight creation which other neural systems center around. CNN

or Convolutional Neural Networks use pooling layers, which are the layers, situated following CNN statement.

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 674

Page 2: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

It takes the contribution from the client as an element map that leaves convolutional organizes and readies a

dense element map. Pooling layers helps in making layers with neurons of past layers.

A pooling layer is another structure square of a CNN. Its ability is to sensibly diminish the spatial size

of the depiction to reduce the proportion of parameters and estimation in the framework. Pooling layer chips

away at every part map unreservedly. The most notable approach used in pooling is max pooling. The totally

related (FC) layer in the CNN addresses the segment vector for the data. This segment layer holds information

that is fundamental to the data. A softmax layer, allows the neural framework to run a multi-class work. To lay it

out simply, the neural framework will by and by have the alternative to choose the probability that the canine is

in the image, similarly as the probability that additional things are joined moreover.

II. LITERATURE REVIEW

Amara, Algergawy J, Bouaziiz, B. [1] says plant illnesses are significant factors as they bring about

genuine decrease in quality and amount of farming items. In this way, early discovery and conclusion of these

maladies are significant. To this end, the creator propose a profound learning-based methodology that

mechanizes the way toward arranging banana leaves ailments. Specifically, the examination utilize the LeNet

engineering as a convolutional neural system to characterize picture informational indexes. Its side effects start

by microscopic, chlorotic spots and it at that point forms into meager dark colored streaks that are limited by

leaf veins.

Figure 1. Architecture of Deep Convolutional Neural Network

Boukhalfa, M., Brahimi, k., and Moussaoui, A[4] says a few investigations have put resources into

the Machine Learning classifiers to shield plants from ailments by preparing Leaves Images. The acquired

outcomes are empowering, arriving at 99.18% of precision, which our performs drastically shallow models, and

they can be utilized as a down to earth device for ranchers to secure tomato against infection. Tomato possesses

a conspicuous spot in the Algerian rural economy.

Ferentinos, K.P.[7] says, Convolutional Neural Network models were created to perform plant

infection identification and determination utilizing basic leaves pictures of solid and unhealthy plants, through

profound learning strategies. Also, on account of huge scope developments, the framework could be joined with

independent horticultural vehicles, to precisely and auspicious find phytopathological issues all through the

development field, utilizing constant picture catching. All these are, obviously, substantial under the condition

that the framework could accomplish elevated levels of execution in recognizing and diagnosing explicit

illnesses, in actuality, conditions (i.e., in the development field), and that it could be worked through a proper,

simple to-utilize, and easy to use portable application for the particular instance of wheat plants.

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 675

Page 3: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

K.P, Ferentinos [8] says, Convolutional Neural Network models were made to perform plant

contamination area and assurance using essential leaves pictures of sound and undesirable plants, through

significant learning techniques. Getting ready of the models was performed with the use of an open database of

87,848 pictures, containing 25 one of a kind plants in a ton of 58 specific classes of [plant, disease] mixes,

including strong plants. All these are, clearly, generous under the condition that the structure could achieve huge

degrees of execution in recognizing and diagnosing express diseases, taking everything into account, conditions

(i.e., in the advancement field), and that it could be worked through a legitimate, easy to-use, and simple to

utilize compact application for the specific case of wheat plants.

Liu, B., He,D.,Zhang, Y ., and Li, Y[13] recognized Mosaic, Rust, Brown spot, and Alternaria leaf

spot are the four basic kinds of apple leaf sicknesses. Early conclusion and exact distinguishing proof of apple

leaf maladies can control the spread of contamination and guarantee the sound improvement of the apple

business. The current research utilizes complex picture preprocessing and can't ensure high acknowledgment

rates for apple leaf maladies. The creator proposes an exact distinguishing approach for apple leaf sicknesses

dependent on profound convolutional neural systems. This examination demonstrates that the proposed

profound learning model gives a superior arrangement in illness control for apple leaf sicknesses with high

exactness and a quicker union rate, and that the picture age system proposed the exploration can improve the

heartiness of the convolutional neural system model.

Figure 2. AlexNet Inception

Lu, Y .,Zeng, N ., Yi, S ., Liu, Y., Zhang, Y[14] says the programmed recognizable proof and

analysis of rice illnesses are profoundly wanted in the field of horticultural data. Profound learning is a hot

research subject in design acknowledgment and AI at present, it can adequately take care of these issues in

vegetable pathology.The creator propose a novel rice maladies distinguishing proof strategy dependent on

profound convolutional neural systems (CNNs) methods.The conventional strategy for identifying rice

infections requires bunches of specialists' understanding and information. With the advancement of PC and web

innovation, ranchers can look through the rice maladies pictures database or counsel the plant pathologists to

pass judgment on rice.

Mohanty, S. P., Hughes, D. P., &Salathe, M[15], says that crop ailments are a significant danger to

nourishment security, yet their quick recognizable proof stays troublesome in numerous pieces of the world

because of the absence of the vital infrastructurie. In the creating scene, in excess of 80 percent of the agrarian

creation is produced by smallholder ranchers, and reports of yield loss of over half because of vermin and

ailments are normal. Besides, the biggest division of hungry individuals (half) live in smallholder cultivating

family units, making smallholder ranchers a gathering that is especially powerless against pathogen-inferred

disturbances in nourishment supply.

Jose G.M.Esgario, et al[11] explaine the biotic pressure comprises of harm to plants through other

living creatures. Effective control of biotic specialists, for example, nuisances and pathogens is firmly identified

with the idea of horticulture supportability. Horticulture maintainability advances the improvement of new

advances that permit the decrease of natural effects, more prominent availability to ranchers and subsequently,

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 676

Page 4: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

increment on efficiency. The utilization of PC vision with profound learning strategies permits the early and

address distinguishing proof of the

pressure causing operator.

Figure 3. Potato defect classification using CNN

Paper Title

Network

Dataset

No. of

training

images

No. of

testing

images

Pros

Cons

Accur-

acy

A Deep learning

Based Approach for

Banana leaf

Classification

CNN

(LeNet)

Plant

village

3700

2057

Food served

on the leaves

absorbs the

polyphenols.

Nutrient

imbalance,Sl

eepiness.

97%

Deep Learning for

tomato Diseases:

Classification and

symptoms

visualization

CNN

(AlexNet,

GoogLeNet

)

Plant

village

800

50

It helps to

protect from

cancer,

maintain

diabetes.

In large

amount of

tomato

leaves cause

poisoning.

99%

Automated

Identification of

Norther leaf

Blight-Infected

Maize Plants from

Field Imagery

Using Deep

Learning

CNN

(pipeline)

Corn

Images

(own)

100

421

Consumed

widely as

food and

feed, not

toxic, not

allergenic.

Stability of

Proteins in

seed may

lead to

Persistence

97%

Deep Learning

Models for Plant

Disease Detection

and Diagnosis

CNN

(Several)

Plant

village

70300

17548

It take less

time,less

efforts and

become

more

accurate.

It take more

steps for

becoming

good leaves.

99%

A Robust Deep

Learning Based

Detector for Real

Time Tomato

Plant Disease and

Pests Recognition

Sensors.

CNN

(Several)

Tomato

Images

(own)

43398

100

It helps to

protect from

cancer,

maintain

diabetes.

In large

amount of

tomato

leaves cause

poisoning.

83%

Table 1: Comparison of different plant leaf diseases

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 677

Page 5: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

Identification of

Apple Leaf

Diseases Based on

Deep

CNN

CNN

(AlexNet)

Apple

Images

(own)

10888

2801

It is good for

heart.They

are linked to

lower risk of

Diabetes

It is based

on calories,

loose stools,

stomachache

98%

Identification of

Rice Diseases

Using Deep

Convolutional

Neural Network

CNN

Rice

Images

(own)

500

85

Adding a lot

of rice could

lead to

chronic

diseases

It contains

more

calcium and

iron

95%

Using Deep

Learning for

Image Based

Plant Disease

Detection

CNN

(AlexNet,

GoogLenet

)

Plant

Village

41112

54306

It take less

time,less

efforts and

become

more

accurate.

It take more

steps for

becoming

good leaves.

99%

Potato Disease

Classification Using

Convolutional

Neural Network

CNN

(AlexNet,

GoogLeNet

,VGG-16)

Potato

Images

(own)

168

115

It contain

fiber and

nutrients,

they are very

satiating and

versatile

It decrease

the risk of

heart disease

95%

Deep Learning for

Classification and

Severity

Estimation of

Coffee Leaf

Image Biotic Stress

CNN

(Several)

Coffee

Images

(own)

1052

210

It boost

physical

performance

,loss weight

burns fat.

It is toxic,

cause

insomnia.

96%

III.EXISTING SYSTEM

The exploration works for a point by point conversation of 65 models, just as quantitative examination

of various strategies in the two significant research headings: Human Fixation Prediction and Deep Learning

Computation. In the Deep Learning Object Prediction uses low level preparing to decide the differentiation of

picture areas to their environment, use include properties, for example, force, shading, and edges.

Deep Learning Computation Object Detection is basic, naturally conceivable, and simple to parallelize to

multi-scale differentiate by directly joining contrast in a Gaussian picture pyramid. Most as of late at the same

time model neighborhood low-level pieces of information, worldwide contemplations, visual association rules,

and elevated level highlights to feature Guided ailment protests alongside their specific circumstances. Such

strategies utilizing nearby differentiation will in general produce higher Deep Learning Computation esteems

close to edges rather than consistently featuring Guided infection objects. The normal Deep Learning

Computation esteems inside picture portions created by mean-move division, and afterward find Guided ailment

questions by distinguishing picture sections that have normal Deep Learning Computation over a limit that is set

to be double the mean Deep Learning Computation estimation of the whole picture. Leaf shape report the key

drawback in leaf ID. Up to now, a few structure alternatives are attracted out to clarify the leaf structure. Be that

as it may, there's no right application to group the leaf once catching its picture and distinguishing its

characteristics, nonetheless. In plant leaf arrangement, leaf is classed and upheld for its totally extraordinary

morphological choices. Some of the grouping procedures utilized are

Fuzzy logic

Principal component Analysis

k-Nearest Neighbor Classifier

Drawbacks

1. Less Color code density.

2. Artifacts may appear.

3. Segmentation accuracy not proper.

4. Edges not clear.

5. Inaccurate results in extraction high density images.

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 678

Page 6: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

IV.PROPOSED METHODOLOGY

The examination is to distinguish the plant sicknesses and give the answers for recuperate from the leaf

ailments. In the proposed framework the exploration is to giving an answer for recoup from the leaf infections

and furthermore show the influenced piece of the leaf by picture preparing method. Fig no.1 shows framework

engineering of proposed plant leaf ailment location framework. First the pictures of different leaves are obtained

utilizing high goals camera in order to improve results and effectiveness. Computerized camera or comparative

gadgets are utilized to catch pictures of leaf of various sorts, and afterward those are utilized to recognize the

influenced zone in leafs. At that point picture preparing procedures are applied to those pictures to draw out the

helpful highlights which will be required for additional examination. At that point various sorts of picture

handling methods are applied to process the pictures, to get extraordinary and helpful highlights required to

dissect later.

In all the methodologies depicted in the exploration were the picture of the plant leaves is resized to

256 x 256 pixels for the expectation of ailments in the leaves. Over the analyses, there are three unique

renditions of the Whole PlantVillage datasets. First beginning with the PlantVillage dataset as it is,in shading, at

that point explore different avenues regarding a dim scaled rendition, thus evacuating all the additional

foundation data which may can possibly present some natural inclination in the dataset because of the

regularized procedure of information assortment. Division was computerized by the methods for a content tuned

to perform well on our specific dataset. The quantities of parameter are utilized for preparing, approval and

testing process. To get a feeling of approaches will perform on new inconspicuous information, and furthermore

to monitor if any of the methodologies are over fitting, the run of all trials over an entire scope of train-test set

parts, specifically 80-20(80% of the entire dataset utilized for preparing, and 20% for testing), 60-40(60% of the

entire dataset utilized for preparing, and 40% for testing), 50-50(50% of the entire dataset for preparing, and half

for testing), 40-60 (40% of the entire dataset for preparing, and 60% for testing), 20-80(20% of the entire dataset

for preparing, and 80% for testing). For the forecast of plant infections the Feature Extraction and Classification

are utilized. The element extraction model is where the system figures out how to distinguish diverse elevated

level highlights from the info pictures. It comprises of an arrangement of convolution, pooling and completely

associated layer. Also these layers are clarified in the module depiction. In the element extraction model five

convolution layer and two pooling layers are utilized. The arrangement model the completely associated layers

where every neuron gives a full association with all took in include maps gave from the past layer in the

convolution neural system. These associated layers depend on the softmax enactment work so as to process the

classes scores. The contribution of the softmax classifier is a vector of highlights coming about because of the

learning procedure and the yield is a likelihood that a picture has a place with a given class. In the grouping

model completely associated layer and softmax layer is utilized to anticipate the plant sickness wether it is a

solid leaves or influenced by any infections. Another system for Deep Learning Computation dependent on

spectural area whether the calculation utilizes the band-pass separating in Fourier transform(FT) space with a

few data transfer capacities that can speak to mindful locales on the pictures. The higher data transmissions at

higher recurrence edges or corners can be distinguished on the pictures. The works, surface portrayals are given

higher loads to make consistency on the recognized Guided infection areas. Nearly the precision of the Plant

Disease Prediction is expanded to 99.5% utilizing Deep Learning technique.

ADVANTAGES

1. Object of interest image segmentation.

2. Adaptive compression.

3. Object recognition.

4. Content aware image editing.

5. Object level image manipulation.

Bacterial Diseases

A bacterial infection is for the most part alluded as the "Bacterial leaf spot". It is started as the little,

yellowgreen injuries on youthful leaves which normally observed as disfigured and turned, or as dull, water-

doused, oily - showing up sores on more established foliage.

Viral Diseases

All popular illness presents some level of decrease underway and the life of infection contaminated

plants is generally short. The most accessible side effects of infection contaminated plants are every now and

again show up on the leaves, yet some infection may cause on the leaves, foods grown from the ground. The

Viral malady is extremely hard to examine. Leaves are viewed as wrinkled, twisted and development might be

modest because of the infection.

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 679

Page 7: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

Fungal Diseases

Contagious infection can impact the Contaminated seed , soil, yield, weeds and spread by wind and

water. In the early on compose it appears on lower or increasingly prepared gets out as water-drenched, dark

green spots. A while later these spots are dark and by then white parasitic improvement spread on the

undersides. In fleece development yellow to white streak on the upper surfaces of increasingly prepared gets out

occurs. It spreads outward on the leaf surface making it turn yellow.

Table 2. Comparison of Deep Learning and Machine Learning

S.NO DEEP

LEARNING

MACHINE LEARNING

1. A subset of machine learning

based on neural networks that

permit a machine to train itself

to perform a task.

Machine learning is the practice of

getting machines to make

decisions without being

programmed.

2. To make machines learn

through data so that they can

solve problems.

To build neural networks that

automatically discovers patterns

for feature detection.

3. Once they are implemented,

the algorithms are usually self-

directed for the relevant data

analysis.

The various algorithms are

directed by the analysts to

examine the different variables in

the datasets.

4. The output can be anything

from a score, an element, free

text or sound,

The output is usually a numerical

value, like a score or a

classification.

V. MODULES DESCRIPTION

1. Image Preprocessing

2. Convolutional Neural Network

3. Pooling Layer

4. Fully Connected Layer

Image preprocessing

Image pre-processing is used to increase the quality of image necessary for further processing and

analysis. It includes color space conversion, image smoothing and age enhancement. The nature of info picture

is accomplished by expelling undesired twisting from the picture.

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 680

Page 8: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

Image enhancement is performed to increase the contrast of image. Image clipping is done to get an

interested region . Smoothing channel is utilized for picture smoothing.

Convolutional Neural Network

A convolutional Neural Network is a kind of fake neural system utilized in picture acknowledgment

and preparing that is extraordinarily intended to process pixel information. Convolutional Neural Network are

amazing picture handling, man-made brainpower that utilization profound figuring out how to perform both

generative and engaging assignments, frequently utilizing machine vison that incorporates picture

acknowledgment, alongside recommender frameworks and common language preparing. A CNN utilizes a

framework much like a multilayer perceptron that has been intended for diminished preparing necessities. The

layers of a CNN comprise of information layer, yield layer and concealed layer that incorporates various

convolutional layers, pooling layers, completely associated layers and standardization layers.

Figure 4. Detection of plant disease using image processing

Pooling Layer

The max-pooling operation was adopted after the convolutional layers. The first convolutional layer C1

is composed of 32 feature maps which are connected to all of image patched of 16 × 16 through filters of size 5

× 5 and stride of one pixel. The size of the feature maps generated in this layer was 14 × 14. The second layer

P1 is max-pooling with kernel size of 2 × 2 and stride of 2. Layer C2 took the output of P1 as input with size 7 ×

7. We again used the same convolutional operation to obtain 32 feature maps with the size of 7 × 7. Then

pooling operation is applied, in Layer P2, which results in maps with size 2 × 2. The Fifth layer is fully-

connected layer F6 was applied in the top of CNNs in order to discover the relationships between high-level

features obtained from previous layers, where there were 64 neurons in this layer. The final layer contained two

units fully connected with the layer F6, one neuron activated by softmax regression produced a value between 0

and 1 which can be interpreted as the probability of the pixel centered at the patch being Leaf disease present or

not.

Pooling layer give a way to deal with down testing highlight maps by outlining the nearness of

highlights in patches of the component map. The two regular pooling strategies are normal pooling and max

pooling that diagram the normal nearness of a component and the most actuated nearness of an element

separately. Pooling is required to down example the location of highlight in include maps. Max pooling is an

example based discretization process. The goal is to down-example an info portrayal, decreasing its

dimentionality and taking into consideration suspicions to be made about highlights contained in the sub-

districts binned. Normal pooling layer performs down-testing by jumping the contribution to rectangular pooling

locales and registering the normal estimations of every district. Normal pooling layer includes ascertaining the

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 681

Page 9: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

normal for each fix of the element map. This implies every 2 x 2 square of the component map is down

examined to the normal incentive in the square.

Fully Connected Layer

Fully connected layer is a neural networks are those layers where all the contributions from one layers

are associated with each initiation unit of the following layer. In most mainstream AI models, the last scarcely

any layers are completely associated layers which orders the information extricated by past layers to shape the

last yield.

Fully connected layers are a basic part of convolutional neural system which have been demonstrated

effectively in perceiving and grouping pictures for PC vision. The CNN procedure starts with convolution and

pooling, separating the picture into highlight and investigating them freely.

The proposed system is tested with three different patch size 16×16, 24×24 and 32×32. The results of

these models were presented in the next section. As the complexity of the network increase, the execution time

also increases. The key challenge is to balance the accuracy and the execution time. We use the limited feature

map, if we increase the number of features maps the accuracy will be improved. These DCNNs architectures are

showed in Table 1 was used to implement the DCNNs model.

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 682

Page 10: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

Table 3. Three DCNN Architectures

Architecture

/ Layer

Architecture 1 Architecture 2 Architecture 3

Layer 1

C1 : Convolutional Layer

with 32 Feature Map,

5×5 size and 1 Stride

C1 : Convolutional

Layer with 32 Feature

Map, 5×5 size and 1

Stride

C1 : Convolutional Layer

with 32 Feature Map, 5×5

size and 1 Stride

Layer 2

P1: Max Pool with 2×2

and stride 2 (with

padding)

P1:Avg Pool with 2×2

and stride 2 (without

padding)

P1:Max Pool with 2×2

and stride 2 (with padding)

Layer 3

C2 : Convolutional Layer

with 32 Feature Map,

3×3 size and 1 Stride

C2 : Convolutional

Layer with 32 Feature

Map, 3×3 size and 1

Stride

C2 : Convolutional Layer

with 32 Feature Map, 3×3

size and 1 Stride

Layer 4 P2: Max Pool with 2×2

and stride 2

P2: Avg Pool with 2×2

and stride 2

P2: Max Pool with 2×2

and stride 2

Layer 5 Fully Connected Layer

C3 : Convolutional

Layer with 64 Feature

Map, 3×3 size and 1

Stride

C3 : Convolutional Layer

with 64 Feature Map, 3×3

size and 1 Stride

Layer 6 Softmax Classifier Fully Connected Layer Fully Connected Layer

Layer 7 - Softmax Classifier Softmax Classifier

VI. RESULT

Almost in the above work accuracy of the Plant diseases prediction is 99% and resolutions of the image

on edges is not clear for the prediction. Nearly in the above work the exactness of the Plant maladies forecast is

99% and the goals of the picture on the edges can't for the expectation. So for right now precision is improved to

99.5% and the picture goals on the edges is unmistakably related to the Fourier Transform (FT) utilizing the

Deep Learning Convolutional Neural Network. The prepared and tried pictures were stretched out before the

work for the best outcome for the forecast of leaf maladies. So for the Prediction precision is plainly seen with

the correlation of before work and the work which is improved for the best outcome utilizing the Deep Learning

Convolutional Neural Network.

Yellow leaf curl virus

It is a Yellow leaf curl virus which influences the leaf spots over 380 have species of plant. It can too

influence leaf spots, rots, blight and other plant parts.

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 683

Page 11: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

Remedies of Yellow leaf curl virus which shown following screenshot.

Bacterial Spot

Bacterial Blight is characterized by small, pale green spots or streaks appeared as water-soaked. The

lesions will expand then appear as dry dead spots. It may extend until the full length of the leaf.

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 684

Page 12: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

Remedies of Bacterial Spot Leaf which shown following screenshot.

Late Bright Leaf

Late Bright Leaf is characterized by small, pale ash spots or streaks appeared as water-soaked. The

lesions will expand then appear as dry dead spots.

Remedies of Late Bright Leaf which shown following screenshot.

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 685

Page 13: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

Accuracy

The proposed DCNN architectures are compared and performance of each architecture evaluated by

accuracy. Accuracy is referring to the closeness of a measured value to an actual value. And also precision is

independent of accuracy, which calculated by,

Accuracy =true positive + true negative

true positive + true negative + false positve + false nagtive (4)

Table 4 shows the accuracy of the proposed DCNN architectures. Each architecture composed of different

number of layer and it is evaluated with three different patch sizes (16×16), (24×24), (32×32). The architecture

3 yields the higher accuracy with the patch size of 24 × 24.

Table 4. Comparison of Accuracy for Proposed DCNN architectures

DCNN

Architecture Patch Size Accuracy (%)

DCNN

Architecture 1

16×16 74.68

24×24 75.76

32×32 70.91

DCNN

Architecture 2

16×16 71.11

24×24 73.25

32×32 73.16

DCNN

Architecture 3

16×16 82.42

24×24 82.55

32×32 80.02

Figure 5. Accuracy of Proposed DCNN Architectures

Table 4 shows the results comparison of proposed DCNN and the existing techniques includes SVM

and ANN in terms of Precision, Recall, and Dice Score Coefficient. Here 150 samples are taken as input for

16*16

24*24

32*32

16*16

24*24

32*32

16*16

24*24

32*32

DCNNArchitecture

1

DCNNArchitecture

2

DCNNArchitecture

3

5055606570758085

DCNN Architectures

Acc

ura

cy (

%)

Accuracy (%)

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 686

Page 14: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

evaluating the performance and result shows that the DCNN achieves higher true positives. The precision, recall

and DSC give substantially more relevant results for DCNN architecture.

Table 5. Dataset for image classification of plant disease.

Class

No. of

original

images

Total no. of

images:

Original &

augmented

No. of images

from dataset

used for

validation

1.Healthy Leaf

565

4523

331

2.Pear,

Cherry

265

2124

152

3.Peach, Powdery

mildew

108

1296

90

4. Peach, Taphrina

deformans

152

1552

156

5. Apple, pear,

Erwinia amylovora

232

2368

205

6. Apple, pear,

venturia

183

2200

151

7.Apple, powdery

mildew

120

1440

118

8. Apple, Rust

163

1960

163

9. Pair,

Gymnosporangium

sabinae

267

2142

185

10. Pair, Gray leaf

spot

122

1464

198

11. Grapevine, wilt

287

2300

114

12. Grapevine,

mites

250

2000

230

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 687

Page 15: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

13. Grapevine,

powdery mildew

237

1900

183

14. Grapevine,

downy mildew

297

2376

201

15. Background

images

1235

1235

112

Total

4483

30880

2589

Figure 5. Prediction Accuracy

VII.CONCLUSION

The order of plant maladies utilizing advanced pictures is extremely testing. Profound learning

procedures, and CNNs specifically, are apparently prepared to do appropriately addressingmost of the

specialized difficulties related to plant sickness characterization. Then again, dataset constraints as far as both

number and assortment of tests despite everything forestall the rise of really thorough frameworks for plant

infection grouping. A few endeavors are in progress towards building progressively agent databases, and

information sharing is continuously turning out to be regular practice, yet the information accessible is as yet

constrained. The arrangement proposed in this examination can not just increment the size of picture datasets

fundamentally, yet can likewise expand the assorted variety of the information, as the normal inconstancy inside

each picture is in a roundabout way considered by the region into littler areas. This methodology likewise has a

few weaknesses, however it obviously prompts progressively solid outcomes in a setting of restricted

information accessibility.

REFERENCES

[1] Amara, J., Bouaziiz, B., &Algergawy, A. (2017). A deep learningbased approach for banana leaf diseases

classification. In Lecture notes in informatics (LNI) (pp. 79e88).

[2] Barbedo, J. G. A. (2013). Digital image processing techniques for detecting, quantifying and classifying

plant diseases.SpringerPlusi, 2, 660 in Agriculture, 153, 46e53.

[3] Bengio, Y. (2012). Deep learning of representations for unsupervised and transfer learning. Proceedings of

the Workshopon Unsupervised and Transfer Learning, 27, 17e37.

[4] Brahimi, M., Boukhalfa, K., & Moussaoui, A. (2017). Deep learning for tomato diseases: Classification and

symptomsvisualization. Applied Artificial Intelligence, 31, 299e315.

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 688

Page 16: AUTOMATIC DETECTION AND CLASSIFICATION OF PLANT LEAF ...xajzkjdx.cn/gallery/67-may2020.pdf · neural system is that CNN accepts contribution as a two-dimensional cluster and control

[5] Crosas, M. (2011). The dataverse network®: An open-source application for sharing, discovering and

preserving data. D-LibMagazine, 17(1).

[6] DeChant, C., Wiesner-Hanks, T., Chen, S., Stewart, E. L., Yosinski, J., Gore, M. A., et al. (2017).

Automated identificationof northern leaf blight-infected maize plants from field imagery using deep

learning.Phytopathology, 107, 1426e1432.

[7] Ferentinos, K. P. (2018). Deep learning models for plant disease detection and diagnosis. Computers and

Electronics inAgriculture, 145, 311e318.

[8] Fuentes, A., Yoon, S., Kim, S. C., & Park, D. S. (2017). A robust deep-learning-based detector for real-time

tomato plantdiseases and pests recognition.Sensors, 17, 2022.

[9] Hughes, D. P., &Salathe, M. (2015). An open access repository of images on plant health to enable the

development of mobiledisease diagnostics.arXiv, 1511, 08060.

[10] Irwin, A. (2002). Citizen science: A study of people, expertise and sustainable development (1st ed.). UK:

Routledge, Abingdon-on-Thames.

[11] Jose G.M. Esgario, Renato A. Krohling, Jose A. Ventura(2019). Deep Learning for Classification and

Severity estimation of Coffee Leaf Biotic Stress. arXiv: 1907.11561v1.

[12] Kamilaris, A., &Prenafeta-Boldu´ , F. X. (2018). Deep learning in agriculture: A survey. Computers and

Electronics in Agriculture,147, 70e90

[13] Liu, B., Zhang, Y., He, D., & Li, Y. (2018). Identification of apple leaf diseases based on deep

convolutional neural networks.Symmetry, 10. Article 11.

[14] Lu, Y., Yi, S., Zeng, N., Liu, Y., & Zhang, Y. (2017). Identification of rice diseases using deep

convolutional neural networks.Neurocomputing, 267, 378e384.

[15] Mohanty, S. P., Hughes, D. P., &Salathe, M. (2016). Using deep learning for image-based plant disease

detection.Frontiers inPlant Science, 7.Article 1419.

[16] Oppenheim, D., &Shani, G. (2017). Potato disease classification using convolution neural networks.

Advances in AnimalBiosciences: Precision Agriculture, 8, 244e249.

[17] Sansone, S.-A., Rocca-Serra, P., Field, D., Maguire, E., Taylor, C., Hofmann, O., et al. (2012). Toward

interoperable biosciencedata.Nature Genetics, 44, 121e126.

[18] Silvertown, J. (2009). A new dawn for citizen science.Trends in Ecology and Evolution, 24(9), 467e471.

[19] Wilkinson, M. D., Dumontier, M., Aalbersberg, I. J., Appleton, G., Axton, M., Baak, A., et al. (2016). The

FAIR Guiding Principlesfor scientific data management and stewardship. Scientific Data, 3, 160018.

Journal of Xi'an University of Architecture & Technology

Volume XII, Issue V, 2020

ISSN No : 1006-7930

Page No: 689