2021 International Conference on Emerging Smart Computing and Informatics (ESCI)

AISSMS Institute of Information Technology, Pune, India. Mar 5-7, 2021

Smart Agricultural Robot for Spraying Pesticide with

Image Processing based Disease Classification
2021 International Conference on Emerging Smart Computing and Informatics (ESCI) | 978-1-7281-8519-4/20/$31.00 ©2021 IEEE | DOI: 10.1109/ESCI50559.2021.9396959

Technique
N. R. Dhumale P.C. Bhaskar
Dept. of E & TC Engg. Electronics Technology Department,
Sinhgad College of Engineering, Shivaji University,
Pune, India Kolhapur, India
[email protected] [email protected]

Abstract— The excellent development of crops decides the

different parameters like worth, store consumption moreover
have afforded numerous proposals and imminent for
final money-making profits of farming output. On the other assessment support and to put into practices for farmers,
hand, there are important factors that are required for the making sure the effectiveness of agricultural manufacturing
development of agriculture harvesting, which consist of [2]. Therefore, CVT will be progressively used to the area of
macronutrients, less important nutrients, micronutrients, as well Agricultural Automation (AA) in addition to progressively
as pesticides. Conventionally, the checking of all the grain, fruit, encourage the growth of agriculture to the period of smart
vegetables, etc development mostly depends on an individual cultivation.
human verdict and is not well-timed or exact. Maintain regular This paper presents an automatic disease classification and
surveillance over crop is an important facet of precision pesticide controller based on an IP Technique (IPT). Chapter
agriculture that uses facts at various crop growth phases. This
paper presents an automatic disease classification and pesticide
2 of this paper set ups the growth and functions of CVT in the
controller based on an Image Processing (IP) technique. Having area of AA. Various regions are engaged in this part. Chapter
precise information about the disease for classification and to 3 gives the motivation and objectives of this work. Chapter 4
spray pesticides is very much supportive in enhancing the proposes the application views of CVT for Smart Agricultural
production efficiency. Compared with manual processes, the Robot (SAR) for Spraying Pesticide with IP based Disease
proposed automatic disease classification and pesticide controller Classification Technique. Chapter 5 and Chapter 6 give
system can identify the disease in crops and can present a implementation of SAR and result-discussion respectively.
consistent and precise basis for suitable regulation. Chapter 7 concludes with future development of SAR.
Keywords — Crops disease classification, pesticide controller,
Image processing, robot.
II. LITERATURE SURVEY
I. INTRODUCTION The Crop Diseases Prevention and Control (CDPC), pests
In modern era, agriculture has taken part in a main function along with hoes are the main makers in the elevated-class and
in the worldwide market. Since the people keep on to increase, litter-without charge agricultural manufactured goods as well
urbanization will show the way in the direction of slow fall in as reaching high yields [3]. Manufacturing complete utilization
the region of civilized soil, along with the difficulty on the of complete farming measures to fast as well as precisely
farming scheme will keep on to enlarge. The requirement of diagnose the incidence of bugs as well as infections [4] in land
useful along with secure farming all the grain, fruit, that is used for farming as well as to robotically and precisely
vegetables, etc making techniques is increasing. Conventional guesstimate the harshness of infections is dangerous in CDPC
Agricultural Organization Techniques (CAOT) must be along with the drop of to produce or provide crops, profits or
accompaniment by novel intellection with powerful expertise results failures [5]. In the conventional organization
also enhanced information as well as communication techniques for Agricultural Plant Protection (APP), there are
equipments in the direction of step up the increase in difficulties for example a need of relative awareness, reduced
agricultural efficiency with added accuracy, thereby precision and deprived suitability [6]. At present, these
supporting the advance of high superiority in addition to high- techniques trust something to work well additional on physical
yield farming [1]. In the happening or done before the usual or organization. For practitioners, there are superior necessities
expected time, Computer Vision Techniques (CVTs) have for professionalism. It is hard for these techniques to get
develop into significant instruments in farming process, and universalization, and it is impractical for them to be executed in
their utilization has really enlarged. Skilled as well as real time. It is demanding to reduce the crop injure caused by
intelligent methods based on CVT algorithms are flattering a disease, but in the course of the application of CVT, the APP
general component of COAT, and CV-based agricultural and control measures have been significantly enhanced, as well
automation system is more and more used in agriculture to as the skill to APP has been very much enhanced. APP at
enhance production and effectiveness. By means of the dangerous times can decrease losses, boost effectiveness and
technology improvement of improvements, progressions, encourage sustainable agricultural growth [7]. The Disease
modernizes, reviews, alters or adjustments, the capability has Detection and Classification (DDC) in farmland is an essential
been really enhanced, and the advances in resource capability situation for integrated pest management. At present, farmers

978-1-7281-8519-4/21/$31.00 ©2021 IEEE 604

Authorized licensed use limited to: San Francisco State Univ. Downloaded on June 19,2021 at 06:42:40 UTC from IEEE Xplore. Restrictions apply.
must initially example the as well as then physically calculate The major characteristic of the Robot is the aptitude to
with recognize disease which is time-consuming, labor categorize the healthy and diseases Images of plants using IP.
demanding and error prone. CVT does several of this work For this a particular reason Android APP based wireless
with extra efficient and correct mode. Investigators have camera which will acquire photos within the turf and examine
completed a lot of hard work on this front. the development according to the tallness, colorization of plant
To examine the condition as well as calculate the numeral of etc is used.
aphids on soybean leaves, Mohammadmehdi Maharlooei et al. For classifying the disease SURF algorithm is used to
[8] applied IPT to execute a lot of experiments on soybean extract various features. Here we are creating database for
sows developed in a orangery. The obtained figures were healthy crop as well as leaf disease. The two classes like leaf
development with MATLAB software to recognize as well as spot and whiteflies are considered. SURF algorithm is used to
compute the amount of louses. The technique is stumpy extract various features. After this when the robot moves in
expenditure and has been by means of scientific experiment field it will take photos and extract the features then MATLAB
verified to have outstanding accuracy in high-quality will compare the live photo with the database. So, based on
illumination situation. The disadvantage of this system is that this we can generate an estimate of percentage of healthy
in rinse conditions, there will be variations in the results. plants and plant disease in a given crop field. We are
Identification based on the phantom uniqueness of the goal is displaying all this data on MATLAB GUI.
additional steady as well as correct. The acquired image are then practiced by using IP method,
the practiced are then transformed into required format for
III. MOTIVATION AND OBJECTIVES further processing as well as given it to the controller unit. In
Having precise information about the disease for the controller unit system is predefined; in line with this
classification and to spray pesticides is very much supportive in system the robot which associated to it was restricted. Robot
enhancing the production efficiency. Compared with manual which has numerous motors is turn on by using the relays.
processes, the proposed automatic disease classification and
pesticide controller system can identify the disease in crops and V. SPEEDED-UP ROBUST FEATURES ALGORITHM
can present a consistent and precise basis for suitable Speeded-Up Robust Features (SURF) algorithm is a scale
regulation. The main objectives of this paper are given below. and rotation invariant robust features detector and descriptor,
x To design and develop robust algorithm to detect various initially proposed by Herbert Bay et al. in ECCV 2006
diseases of plants. conference and published in 2008. SURF is usually applied for
x To design a GUI in MATLAB to display the data in a object tracking, traffic supervise in image sequences, 3D
graphical format. reconstruction or increased actuality applications, etc. All
x To develop an ARM7 based Embedded Board from above applications need real-time image succession processing.
scratch. The present processors not make sure real-time features
x To design irrigation robot which can detect Plant health extraction since the frame rate extremely depends on difficulty
and spray pesticide? of the algorithm, 2D filter range, quantity of features that are
sensed in every scale and descriptor kind that is used to explain
IV. SMART AGRICULTURAL ROBOT each point of significance. The SURF algorithm is supported
A CV-based row supervision technique is shown to direct a by the similar principles and steps as Scale Invariant Feature
machine as it is intended separately to impel in the course of Transform; but particulars in every step are dissimilar. SURF
the row of all the grain, fruit, vegetables, etc in an area is shown in Fig. 2.
consistent with the intended idea of unwrapping manner. The algorithm has three major elements:
Then, the counterbalance and direction angle of the robot x Interest point detection,
platform is sensed in real-time to direct the platform due to
detection of all the grain, fruit, vegetables, etc row using x Local neighborhood description
machine vision. This scheme is essentially grown to apply x Matching
several farming production in a lot of countries, for example
picking, harvesting, weeding, pruning, planting, grafting, SURF utilizes square-shaped filters as an estimator of
agricultural classification, etc., Gaussian smoothing. The sum of the original image within a
Here we are proposing farming independent Robot which rectangle can be assessed rapidly using the basic image,
will capture the states in instantaneous along with make a involving assessment at the rectangle's four corners.
decision which cultivated area is suitable for that exacting The objective of a descriptor is to offer a exclusive and
field. For this, it is examined that the field factors for example, healthy description of an image feature, e.g., by relating the
heat, moisture etc. The detail block diagram is shown in strength distribution of the pixels inside the locality of the point
Fig. 1. of attention. The majority descriptors are thus calculated in a
The major characteristic of the Robot is the capability to local approach; hence a description is acquired for each point of
capture the physical condition of plants by IP. For taking attention identified formerly. By comparing the descriptors
photos contained by the turf as well as examine the obtained from different images, matching pairs can be found.
development according to the height, colorization of leaves etc
Web cam is used. So, supported on this one can make an
approximation of proportion of well plants in a provided crop
turf.

605

Authorized licensed use limited to: San Francisco State Univ. Downloaded on June 19,2021 at 06:42:40 UTC from IEEE Xplore. Restrictions apply.
 Feature Extraction

Input Image
Feature Selection
------------------------

Resize Image x Variation Classifier Training

x Mean
x Trained Classifier Test Image
x
x
x Entropy
Disease
Image Preprocessing Present Grade
? N

Y
Spray Grade I Send Data to
Controller
Spray Grade II
Severity Calculation
Spray Grade

Spray Grade

Fig. 1. Block Diagram of Proposed Method

Computational of Integral Image

Computational of Hessian Matrix Determinant

Normalization the determinant Response

Application of the Non-Maximal Suppression and Threshed

Computational of the Interest Point Orientation

Computational of the Interest Point Descriptor

Fig. 2. Steps in SURF

606

Authorized licensed use limited to: San Francisco State Univ. Downloaded on June 19,2021 at 06:42:40 UTC from IEEE Xplore. Restrictions apply.
 VI. SYSTEM IMPLEMENTATION
Start
In SAR system obtained images from camera coursed in
computer using IPT and the finding of disease is completed and
equivalent signal is set to the microcontroller unit. In the
microcontroller unit programming is used to control the robot
Initialize LCD, UART
to move forward and reverse. According to the forwarded
signal robot will spray pesticides on crop. The circuit diagram
for motor control. Pesticides are sprayed through sprinkler
motor. DC motor is converted into on state and off to enable Read All sensor values through 10 bit ADC
careful spraying of pesticides on crops. The power of spraying
system controlled with the help of ARM controller unit. The
Display sensor values on 16x2 LCD display
obstacles are detected using IR sensors.

VII. ALGORITHM FOR SMART AGRICULTURE ROBOT Send all sensor data to MATLAB
The acquired image are then practiced by using IP
method, the practiced are then transformed into required
format for further processing as well as given it to the
controller unit. In the controller unit system is predefined; Is Plant
in line with this system the robot which associated to it was disease
restricted. Robot which has numerous motors is turn on by No detected? Yes
using the relays.

1. Start.
Go t the next Operate the
2. Initialize LCD, UART.
plant pesticide pump
3. Read All sensor values through 10 bit ADC.
4. Display sensor values on 16x2 LCD display.
5. Send all sensor data to MATLAB. Fig. 3. The flowchart of the SAR system.
6. Plant disease detected then go to next step otherwise go
to Step 9.
7. Operate the pesticide pump.
8. Go t the next plant. Table II
Results of the training and testing accuracy
9. Go to step 3 Naive Bayes Classifier Perceptron rule
Samples
Training Testing Training Testing
Robot which has numerous motors is triggered by with the
relays. According to the control given by the microcontroller 1000 89% 79.25% 89.10% 82.92%
unit these motors are turn on and off. The flowchart of the 820 90% 71.67% 89.98% 75.67%
system is shown in Fig. 3. 700 80% 62.29% 87.29% 68.29%
400 90% 57.76% 89.92% 64.76%
VIII. RESULTS AND DISCUSSION 250 90% 52.29% 8.24% 60.23%
A typical hitch of supervised learning is the getting the
Accuracy 89.60% 64.65% 88.91% 70.37%
representative training data set for disease classes. A
enormous quantity of time as well as payments be able to
accumulate if less quantities of trials are entailed for the Table III
training of classifier. An included advantage is the probable to Training and testing time (sec)
diminish training time of classifier. Training data sets are Samples Maximum Likelihood Classifier Delta Learning Rule
structured along with features. It is experiential that the 1000 1230 9862
accuracy of Maximum Likelihood Classifier and Delta 820 1230 7291
Learning Rule is upmost for 1000 numbers of training
700 1230 5789
samples. If numbers of samples are condensed by 2/3 then the
accuracy is 72.29 % and 78.29 % for Maximum Likelihood 400 1230 9632
Classifier and Delta Learning Rule respectively. Likewise, if 250 1230 2567
numbers of samples are reduced by 1/5 then the accuracy is
62.29 % and 70.23 % for Maximum Likelihood Classifier and
Delta Learning Rule respectively. The detail results are shown
in Table I and Table II. The results of disease classification are
shown in Fig. 4.

607

Authorized licensed use limited to: San Francisco State Univ. Downloaded on June 19,2021 at 06:42:40 UTC from IEEE Xplore. Restrictions apply.
 Original Image of white flies Original Image of Rust

Gray Image Gray Image

YcBcR Image
YcBcR Image

Piosons Thresholding Image Piosons Thresholding Image

Past Image

Denoised Image
Denoised Image
Fig. 4. Results showing disease classification
608

Authorized licensed use limited to: San Francisco State Univ. Downloaded on June 19,2021 at 06:42:40 UTC from IEEE Xplore. Restrictions apply.
 A proposal for combination of images classification and Table IV. Detection time of SAR
robotics is offered at this point to assist improve the idea of Samples Detection time (Sec)
complete farm crop disease control. The implanted course of 930 9682
Smart Agricultural Robot (SAR) for Spraying Pesticide with 750 7861
IPbased Disease Classification Technique is considered like 630 6754
that it can robotically scan a pre-set area. The path for the robot
330 3987
can be position like that an whole field can be robotically
186 2127
surveyed. An investigational study is done to make sure if the
images obtained by the robot can be used for consolidation to
that of a superior farm land region. To check this scheme, robot REFERENCES
is prepared to navigate through sections in a preconfigured [1] Anjna, M. Sood, and P. K. Singh, “Hybrid System for Detection and
passageway and pictures in use are stitched jointly to build up Classification of Plant Disease Using Qualitative Texture Features
entire sections. The accuracy of proposed SAR and detection Analysis,” Procedia Comput. Sci., vol. 167, no. 2019, pp. 1056–1065, 2020,
time is given in Table III and Table IV respectively. doi: 10.1016/j.procs.2020.03.404.
[2] P. K. Sethy, N. K. Barpanda, A. K. Rath, and S. K. Behera, “Image
IX. CONCLUSION Processing Techniques for Diagnosing Rice Plant Disease: A Survey,”
Commercial resolution which have possible to help the Procedia Comput. Sci., vol. 167, no. 2019, pp. 516–530, 2020, doi:
farmers in immediate and composite alternative that can have a 10.1016/j.procs.2020.03.308.
wealthy satisfied of information are desirable. Smart [3] Shivani Machha | Nikita Jadhav | Himali Kasar | Prof. Sumita Chandak,
“Crop Leaf Disease Diagnosis using Convolutional Neural Network,” Int. J.
Agriculture Robot is implemented as an independent Trend Sci. Res. Dev., vol. 4, no. 2, pp. 1056–1058, 2020.
agricultural field robot that facility at a micro and utilizes with [4] R. Patil and S. Kumar, “A Bibliometric Survey on the Diagnosis of Plant
immediate recognition of disease next to controlled spraying of Leaf Diseases using Artificial Intelligence,” Libr. Philos. Pract., vol. 2020,
pesticide. It can also offer a clever consolidation at a macro no. 3, pp. 1–25, 2020.
stage. The precision levels for disease recognition for crops are [5] S. Mishra, R. Sachan, and D. Rajpal, “Deep Convolutional Neural Network
based Detection System for Real-time Corn Plant Disease Recognition,”
found to be suitable. Procedia Comput. Sci., vol. 167, pp. 2003–2010, 2020, doi:
Table. III. Accuracy of SAR 10.1016/j.procs.2020.03.236.
[6] P. Prathusha, K. E. S. Murthy, and K. Srinivas, “Plant Disease Detection
Samples Accuracy Using Machine Learning Algorithms,” pp. 213–220, 2020, doi:
930 79.80% 10.1007/978-3-030-46943-6_25.
750 79.78% [7] K. S. Kalaivani, S. Uma, and C. S. Kanimozhiselvi, “A Review on Feature
630 77.69% Extraction Techniques for Sentiment Classification,” Proc. 4th Int. Conf.
Comput. Methodol. Commun. ICCMC 2020, no. November, pp. 679–683,
330 79.52% 2020, doi: 10.1109/ICCMC48092.2020.ICCMC-000126.
186 78.54% [8] Z. Gao, Z. Luo, W. Zhang, Z. Lv, and Y. Xu, “Deep Learning Application in
Overall accuracy 78.61% Plant Stress Imaging: A Review,” AgriEngineering, vol. 2, no. 3, pp. 430–
446, 2020, doi: 10.3390/agriengineering2030029.

609

Authorized licensed use limited to: San Francisco State Univ. Downloaded on June 19,2021 at 06:42:40 UTC from IEEE Xplore. Restrictions apply.