ISSN (Online) 2581-9429

IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 3, Issue 2, March 2023

Impact Factor: 7.301

Plant Life: An Intelligent Mobile Plant Disease

Diagnostic System using Deep Learning
Dr. Muthu Senthil B1, Mummaneni Sravani2, Konduru Theja Sree3, Krithik Shri M. P.4
Professor, Department of Computer Science and Engineering1
Students, Department of Computer Science and Engineering2,3,4
SRM Valliammai Engineering College, Chennai, Tamil Nadu, India

Abstract: Agriculture is an important sector that considerably enhances a nation's economic development.
Farmers encounter several obstacles every year in order to produce high-quality crops. Plant diseases are
a significant factor in the failure of the harvest. Additionally, pests, specialized illnesses, or natural
disasters cause massive crop losses every year. It raises serious concerns about the application of
sustainable advanced technologies to agricultural problems. The proposed intelligent system's main job is
to detect plant disease using an image captured in the mobile application. It employs the region-based
convolution neural network (fast-RCNN) and pre-trained model generation for the set of datasets and
captured images. To ensure long-term viability, the mobile app is integrated with cloud services. The
proposed system also includes recommendations to help farmers determine whether plant conditions are
healthy or unhealthy, as well asthe recommendations to cure disease.

Keywords: fast R-CNN, plant disease, CNN, Android App

I. INTRODUCTION
India is a major agricultural-producing country that is heavily reliant on food production. Agriculture is also the main
industry and source of wealth in India. Agriculture is also the primary source of employment, food, and income for
many Indians. Good agricultural production is influenced by a variety of things. A few of the variables are the
application of fertilizers, the quantity and distribution of rain, the fertility and condition of the soil, and others. Crop
diseases are the main problem that people encounter. Farmers can lack a thorough understanding of the diseases, which
can result in incorrect disease diagnoses and poor problem-solving techniques. An intelligent mobile plant disease
diagnostic system, on the other hand, has become valuable because of its utility in the early diagnosis and detection of
plant diseases using leaf images, whether healthy or unhealthy. Mobile apps that use artificial intelligence to assist
farmers in identifying plant-affected diseases are one such tool being investigated. In this mobile application, we use
deep learning technology to accurately determine the disease. The region-based convolution neural network model is
being used (fast R-CNN). It is made up of three simple steps: Take a photograph with your camera first. The second
step is to generate a diagnosis report, which will inform the user about the disease and whether the leaf is healthy or
unhealthy. The third step suggests treating the disease with fertilizer.

1.1 Existing System

In previous work, The user application takes a picture of a leaf with the phone's camera. The trained model included in
the "Plant Buddy" app is then used to process the taken image. Based on the precision of the image, the programme
then decides if the plant is healthy or unhealthy. The suggested model performs 10–20% worse when photographs are
taken outside. An image dataset that was taken in a real-world environment can be used to get around this restriction.

II. LITERATURE WORK

A.Uzhinskiyet al [1] To provide a new level of service to the farmers' community, plant disease detection was designed
using modern organisation and deep learning technologies. A mobile application that allows users to send sick plants
photos and text descriptions.

Copyright to IJARSCT DOI: 10.48175/IJARSCT-8927 841

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 3, Issue 2, March 2023

Impact Factor: 7.301

Adedamola O. Adedoja et al [2] The system is powered by a web service that performs diagnostics using the CNN
model. The web service is used to provide photographs of plant leaves taken with the developed mobile app, and the
NASNet-Mobile CNN model is used to identify plant diseases.
Sophia Sanga et al [3] The system is powered by a web service that performs diagnostics using the CNN model. The
web service is used to provide photographs of plant leaves taken with the developed mobile app, and the NASNet-
Mobile CNN model is used to identify plant diseases.
Ayesha Siddiqua et al [4]Weston Three raters evaluated the chosen apps using the system we developed for app
evaluation. Inter-rater reliability is computed with intra-rater reliability to check rater agreements on the ratings and
ensure rating consistency.
ZarreenNaowal Reza et al [5] Using the colour co-occurrence methodology, the resultant feature values from the
segmented area will be gathered for texture analysis. To recognise and categorise the disease, sample values from the
pre-defined database will be compared to the extracted values using a multi-SVM classifier.
Nikos Petrellis et al [6]The user's processing of plant images captured using visible, infrared, and other wavelengths of
light. It is common to utilise lesions or spots in different sections of a plant to determine a disease. The colour, size, and
number of these spots can all aid in determining which disease has killed a plant.
N. Petrellis et al [7] A software for smartphones that can identify plant illnesses. It is based on identifying the disease
signature, which is expressed as a set of guidelines for the colour, shape, and placement of spots, as well as information
about previous weather patterns. The collection of plant diseases supported can be expanded or altered by an
agriculturist using the built programme as an end user thanks to the disease signature format.

III. SYSTEM ARCHITECTURE

In this Proposed system design and development of a mobile application-based plant leaf disease detection and fertilizer
recommended system. In the nutrients model, yield prediction and fertilizer observation are critical for farmers to assess
fertiliser nutrient loss from rain, cyclones, and yieldand other natural disasters. An R-CNN-based deep learning
approach is used to train the model for plant leaf disease.
Project's goal is to develop a mobile application that uses deep learning to diagnose plant diseases. And it’s consisting
of three stages of the layer. User layers are used to upload or capture the image of the plant leaf, algorithm layer is used
to identify whether the plant leaf is healthy or unhealthy. In the android application layer, the final layer displays the
leaf status using the deep learning algorithm of the fast R-CNN technique.
1. This project can help farmers by detecting diseases at earlier stages.
2. Better quality crops can be harvested as the diseases are found easily by the farmers.
3. It gives recommendations on how the affected leaf can be prevented and cured.
4. It is easy to maintain and user-friendly.
The overall goal of the project is to predict diseases on the leaf. The images are divided into 80:20 training and testing
phases. During the training period, the leaf is subjected to the fast RCNN algorithm, which is used to determine the
plant's labels. During the testing phase, we can determine the diseases of a plant using an Android application by
capturing a real-time image or uploading images from the dataset.
The above image depicts the layers available in the system architecture such as Mobile App Layer, Pre-processing
Layer, Deep Learning Model Layer, Database Layer.
Mobile App Layer: This layer is responsible for capturing images of the plant leaves through the mobile camera and
sending them to the pre-processing component for further processing.
Pre-processing Layer: This layer is responsible for cropping, resizing, and enhancing the images to improve the
performance of the deep learning model. It then sends the pre-processed images to the deep learning model component.
Deep Learning Model Layer: This layer is responsible for analysing the pre-processed images using convolutional
neural networks (CNNs) to identify the plant disease.
Database Layer: This layer stores information about various plant diseases, including their symptoms, causes, and
treatments.

Copyright to IJARSCT DOI: 10.48175/IJARSCT-8927 842

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science,, Communication and Technology (IJARSCT)

Volume 3, Issue 2, March 2023

Impact Factor: 7.301

Fig 1. Overall system architecture

Flowchart shows whether the leaf is healthy leaf or unhealthy leaf.

3.1. Modules
3.1.1 Image Acquisition
Selecting the diseased plant is the first step. Next, gather the damaged plant's leaves, take a photo of them, or load an
existing image of them into the computer system. In order to make digital photos useful for the system, image
acquisition is used to process
rocess the images using algorithms.

Copyright to IJARSCT DOI: 10.48175/IJARSCT-8927 843

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 3, Issue 2, March 2023

Impact Factor: 7.301

It eliminates visual distortion and noise. Images from the source folder are read by data producers. To process the
photos, data augmentation is employed. This method improves dataset image quality such that Fast R-CNN models can
be created with them.

3.1.2. Feature Generation

The autonomous learning of the features plays a significant role in feature development for plant disease diagnosis. In
order to identify plant illnesses, shapes, textures, and colours of plant leaf photos are typically used. Using features Fast
R-CNN algorithm, automatically extracting features and detecting plant leaf illness. A quick and effective way to utilise
the features that a neural network that has already been trained has learned is through feature extraction. It defines the
output feature by propagating the input image to a very specific layer of our own (completely linked).

3.1.3. Training the Model

The trained model created in the preceding steps can be used to test real-time images for plant disease detection and
recognition. We train the model using fast R-CNN algorithm.
The RPN and Fast R- CNN are two separate networks that make up the Fast R-CNN architecture. They can all be
trained independently. In contrast, it is possible to create a combined network for Faster R-CNN in which both the RPN
and Fast R-CNN are trained simultaneously.
The fundamental concept is that the convolutional layers used by Fast R-CNN and RPN are identical. Although only
existing once, these levels are utilised by the two networks. It is possible to refer to it as sharing of features or layer.

3.1.4. Testing the Model

The testing models are used to test the plant leaf images to detect and recognize the three pages. It’s listed below,
Admin page, Login page, Prediction page.
 Admin Page: When first opened, the application displays the admin page. Here, there are two buttons: Signup
and Login. The Register button displays the user's name, email address, password, and contact number.
 Login Page: This button directs the user to the login page. Capture or upload the leaf image and Result are the
two buttons present. A previously acquired image from the gallery can also be loaded, or a fresh image can be
taken with the camera. The detected plant disease and the precision of the detection are then displayed after
pressing the "Result" button.
 Prediction Page: Following the image's detection, the solution page is shown, where the disease's therapy is
explained.
 Recommendation Page: It gives the recommendations to cure the disease after detecting the disease. After
this the user can logout from the page.

IV. ALGORITHM
Approach 1: Plant Leaf Disease Diagnosis using Fast RCNN
Input: Infected or healthy leaf image
Output: Class of leaf image
1. Read the input image.
2. Read the image into RGB image.
3. Resize the image.
4. Convert input image into the feature extraction.
5. Apply the Fast RCNN algorithm.
6. Diagnosis the input image.
7. Calculate the image is healthy or unhealthy.
8. Calculate the Evaluation metrics.
9. End

Copyright to IJARSCT DOI: 10.48175/IJARSCT-8927 844

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 3, Issue 2, March 2023

Impact Factor: 7.301

V. RESULTS

Fig 2.Signup page

Fig 3.Upload/ Capture image

Fig 4.Disease prediction page

Copyright to IJARSCT DOI: 10.48175/IJARSCT-8927 845
www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 3, Issue 2, March 2023

Impact Factor: 7.301

Fig 5. Recommendation page

Fig.2 describes the signup page. Where a farmer must provide his details for creating an account. the details include
name, email address, and contact number. To create the account for the first time we need to sign up, then onwards we
can log in to the app directly.
Fig.3 shows uploading or capturing the image, where a user may access both the real-time images and the dataset
images. While capturing the images they have to be present in real, the uploading can be done with the already present
dataset.
Fig.4 describes the disease prediction. after submitting the image, the user can able to notice the disease type and the
fertilizer that needs to be used.
Fig.5 gives the recommendations to cure the disease when the disease is detected.

5.1. Evaluation Metrics

The following evaluation measures are used to evaluate the effectiveness of the FE algorithms and ML models:
 TP: True Positive is the quantity of attack samples that were accurately identified.
The number of benign samples that were appropriately identified as TN: True Negative.
 The number of attack samples that were incorrectly labelled as false positives.
The number of benign samples that were incorrectly labelled as false negatives.
Accuracy: Accuracy is calculated as the ratio of samples that were properly identified to all samples:

+
=
+ + +
Recall: Also referred to as detection rate, is the ratio of attack samples that were correctly categorised to all of the
attack samples:
=
+
False Alert Rate: The ratio of assault samples that were mistakenly classified to benign samples is known as the false
alarm rate.
=
+
F1-Score: F1 Score is the harmonic mean of precision and Detection Rate:
2
1=
Area Under the Curve: The trade-off between the DR and FAR is shown by the area under the Receiver Operating
Characteristics (ROC) curve, or area under the curve.
Copyright to IJARSCT DOI: 10.48175/IJARSCT-8927 846
www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 3, Issue 2, March 2023

Impact Factor: 7.301

Fig 5.1 Fast RCNN Algorithm

Fig 5.2 Inception Algorithm

Fig 5.3 Inception Loss

Copyright to IJARSCT DOI: 10.48175/IJARSCT-8927 847
www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 3, Issue 2, March 2023

Impact Factor: 7.301

Fig 5.4 Fast R-CNN Loss

Algorithm Accuracy
ResNet (Residual Networks) 98.8
LRA 92.3
Inception 85.3
CNN (Convolutional Neural Networks) 94.1

VI. CONCLUSION
In this study, a mobile application for Android has been created to quickly and accurately identify and detect plant
illnesses. The model's high accuracy, which was reached using a huge public dataset of plant photos, demonstrates the
model's acceptance. In this essay, it is suggested that an Android app be created to detect plant leaf disease. Using the
Fast RCNN technique, the disease detection model in this approach analyses a plant leaf image as input before
detecting and identifying the disease. With low-end hardware, this suggested way application also runs and performs
well. The user interface for the application is likewise fairly simple. So, farmers may use the programme pretty well. It
can be used by farmers in remote places to correctly identify plant illnesses and administer the required treatments.

REFERENCES
[1]. Uzhinskiy A, Ososkov G, Goncharov P, Nechaevskiy A. Multifunctional platform and mobile application for
plant disease detection. InCEUR Workshop Proc 2019 (Vol. 2507, pp. 110-114).
[2]. Siddiqua A, Kabir MA, Ferdous T, Ali IB,Weston LA. Evaluating plant disease detection mobile
applications: Quality and limitations. Agronomy. 2022 Aug 8;12(8):1869.
[3]. Petrellis N. A smart phone image processing application for plant disease diagnosis. In2017 6th international
conference on modern circuits and systems technologies (MOCAST) 2017 May 4 (pp. 1-4). IEEE.
[4]. Adedoja AO, Owolawi PA, Mapayi T, Tu C. Intelligent Mobile Plant Disease Diagnostic System Using
NASNet-Mobile Deep Learning. IAENG International Journal of Computer Science. 2022 Mar 1;49(1):216-
31.
[5]. Sanga S, Mero V, Machuve D, Mwanganda D. Mobile-based deep learning models for banana diseases
detection. arXiv preprint arXiv:2004.03718. 2020 Apr 7.
[6]. Nikos Petrellis et al [6] , smart phone image processing application for plant disease diagnosis Mobile
Application for Plant Disease Classification Based on Symptom Signatures

Copyright to IJARSCT DOI: 10.48175/IJARSCT-8927 848

www.ijarsct.co.in
 ISSN (Online) 2581-9429
IJARSCT
International Journal of Advanced Research in Science, Communication and Technology (IJARSCT)

Volume 3, Issue 2, March 2023

Impact Factor: 7.301

[7]. Reza ZN, Nuzhat F, Mahsa NA, Ali MH. Detecting jute plant disease using image processing and machine
learning. In2016 3rd International Conference on Electrical Engineering and Information communication
Technology (ICEEICT) 2016 Sep 22 (pp. 1-6). IEEE.
[8]. Petrellis N. Mobile application for plant disease classification based on symptom signatures. InProceedings
of the 21st Pan-Hellenic Conference on Informatics 2017 Sep 28 (pp. 1-6).
[9]. Siddiqua A, Kabir MA, Ferdous T, Ali IB, Weston LA. Evaluating Plant Disease Detection Mobile
Applications: Quality and Limitations.Agronomy 2022, 12, 1869.

Copyright to IJARSCT DOI: 10.48175/IJARSCT-8927 849

www.ijarsct.co.in