Volume 7, Issue 6, June – 2022 International Journal of Innovative Science and Research Technology

ISSN No:-2456-2165

Android Application on E-commerce & Plant

Disease Identification using Tensor Flow
Shrikant S. Gurav, Snehal A. Patil, Shubhangi L. Patil, Priti P. Desai, Prajakta B. Badkar,
Department of Computer Science & Engineering
Sant Gajanan Maharaj College of Engineering, Gadhinglaj, Kolhapur

Abstract:- In this Digitized world farmers can do day-to- advancements that have brought about genuine financial
day operations using online means such as buy and sell of advantage for the farmers and one in every one of them can
saplings. The Modern world can connect back to the use apps like PlantIP for increasing their reach to customers
motherland using PlantIP app which helps farmers to and selling veggies and fertilizers, small gardening plants and
digitize themself and form a community. To identify and much more using easy customization of selling products.
detect plant diseases, which are a major factor in the loss
of crop yield in agriculture and crop production, we The new era of Agriculture brought lots of new
employ a powerful ML library provided by Google. The advancement in technologies as well as lots of untrack
project uses live weather forecast for a week with updated diseases which hard to keep it up with here comes the use of
GPS tracks using satellite data from Openweathermap. PlantIP the one stop solution
Parts of the distributed run-time system for the plant
disease detector are arranged to execute on user-side To detect any disease stored on cloud and processed
mobile devices and cloud-side centralized servers. The using google Teachable machine [1].
CNN deep learning model and the Intermediate The neural network is formed to optimize the shortest
Representation (IR) model that operate on mobile devices path [2] to guess the disease and finding a cure and
are both described in Layer 1 of the system. The user information to handle disease is provided using PlantIP.
interface is depicted in Layer 2 and created as an
Android app to make it easy for system users (shown in We have used more than thirty-eight types of disease
Layer 3) to engage with the system. which are more common for basic farming and home
gardening yields.
Keywords:- Image processing, image acquisition,
classification, segmentation. We use Open weather map for live updates of weather
in the app as nature is so unpredictable these days, the
I. INTRODUCTION increasing global warming needs proper plans for the week
PlantIP with the approach of image processing without interposition of any preventing measures for farming.
Technology and connecting to Community of agriculture
experts’ know-how or help fellow farmers with their
knowledge and experience: Which will make the PlantIP
Community, the largest network for farmers worldwide.
Where they can talk about new methodologies and get good
yields with our App.

This paper consists of details about the app and how it

helps farmers to reach out to customers as well as form a
community by providing help in live weather tracking and
week planners.
Fig. 1
Parts of the distributed run-time system for the plant
disease detector are arranged to execute on user-side mobile
devices and cloud-side centralised servers. The CNN deep
learning model and the Intermediate Representation (IR)
model that operate on mobile devices are both described in
Layer 1 of the system. The user interface is depicted in Layer
2 and created as an Android app to make it easy for system
users (shown in Layer 3) to engage with the system. literature
survey & use cases

After the recent Lockdown world has turned its clock to

go faster in world of digital media, People are more familiar
and comfortable ordering online goods. Lots of farmers like
trying new things to have more crop yield. The field of
agriculture is in decline, yet there have been a number of Fig. 2

IJISRT22JUN602 www.ijisrt.com 812

Volume 7, Issue 6, June – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 A UI of use case of Open weather map in Plantip App. B. Late Blight Potato Leaf
 We use a scheduler planner to remind important updates
regarding the weather forecast and notify using OneSignal
notification.
 The users can never miss what's important using plantip as
the scheduler is highly customizable for location-based
notifications from user frequently used locations.

Fig. 5: Late Blight potato Leaf

The oomycete microbe Phytophthora infesters, which

resembles a fungus, causes late blight in potato plants. This
unquestionably dangerous disease can affect potato leaves
and tubers at any stage of crop growth.

C. Tomato Bacterial Spot

Fig. 6: Tomato Bacterial Spot

Xanthomonas vesicatoria, Xanthomonas euvesicatoria,

Fig. 3: Plant IP app screenshots Xanthomonas gardneri, and Xanthomonas perforans are the
culprits behind tomato bacterial spot. On contaminated seed
 The E-commerce buy and sell helps direct relations and transplants, which may or may not exhibit symptoms,
between farmers and customers selling 100 of categories these bacterial infections can be introduced into a garden.
including plants, fertilizers, small farming equipment etc.
 The smart suggestions help buying more easily in plantip as Dataset used to train model:
users can perform all operations like review and request
customizations.

II. DISEASE TYPES & CLASSIFICATION

A. Tomato-Early blight
Corn-common rust grey leaf spot blight leaf

Grape-black rot grape-esca leaf blight

Fig. 4: Tomato Early blight Fig. 7

The fungus Alternaria solani is the source of the

widespread tomato disease known as early blight. Almost all
tomato plant parts, including the leaves, stems, and fruits, are
susceptible to it. Although the plants might not perish, they
will likely be weaker and produce fewer tomatoes than usual.
Although it can affect seedlings as well, early blight often
affects older plants.

IJISRT22JUN602 www.ijisrt.com 813

Volume 7, Issue 6, June – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
III. PROPOSED WORK & ML USE

PlantIP with the approach of image processing

Technology and connecting to a network of agriculture
experts’ know-how or help fellow farmers with your
knowledge and experience. Where they can talk about new
methodologies and get good yields with our App [3]. Fig. 9
Parts of the distributed run-time system for the plant First, the user launches the app, selects a plant image, and
disease detector are arranged to execute on user-side mobile then clicks a button. The image is sent to the TensorFlow
devices and cloud-side centralized servers. Lite model included with the app after pressing the button,
which produces a class label. The screen of the device merely
The system's deep learning model and the mobile shows this class label.
device's Intermediate Representation model are both
described in Layer 1. IV. ARCHITECTURE
The user interface is depicted in Layer 2 and created as  STEP 1: Android Application Setup:
an Android app to make it easy for system users (shown in The application requires the user to provide correct
Layer 3) to engage with the system. credentials for the user to register or log in and to provide
the correct permissions for the application. The Android
In the first step of the system we propose, image
application uses Tensor Flow, which is integrated, to
acquisition, we take pictures with a digital camera.
leverage trained models.
Second, to enhance the leaf's damaged region and
 STEP 2: Making a learnt model from a dataset:
remove noise from the provided image, image processing
Using the correct Tensor Flow methods, such as clustering,
techniques like image enhancement and image segmentation
a model is made that has already been pre-trained from a
are applied to the leaf.
collection of plants.
Finally, Teachable Machine's convolution neural
network is used to complete the classification of the sick
leaves. The processes below are repeatedly carried out for
each image in the data collection.

The dataset's images are first imported into memory,

then fed to the CNN model with the proper input sizes. After
the CNN model computes and processes the image's
information, it then outputs the image's classification.

Fig. 8

The above figure represents the architecture of the

deployment of the trained CNN model to the Android app
using TensorFlow Lite.

After creating the CNN model, we first store it in the.h5 Fig. 10: Flowchart
format, which freezes or saves the trained weights along with
the model. Using the TocoConverter tool, we can transform  STEP 3: Datasets for training and testing:
our TensorFlow model into a TensorFlow Lite model, which The TensorFlow comprises information about the plant
can then be included in our Android app. This model can disease and the model developed using this data. The ratio
complete tasks without an internet connection thanks to its of training to testing sets is 10:1, correspondingly. The
mobile device optimization. TensorFlow framework uses the prewritten clustering
algorithm for machine learning. [4].

IJISRT22JUN602 www.ijisrt.com 814

Volume 7, Issue 6, June – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
 STEP 4: Testing data:  Draw the model Metrix graph
The "Identify the illness" option is available in the
application. This further enables the algorithm to forecast
illness on the fresh plant sample biased on the information
supplied by farmers about the plants, if needed.

 STEP 5: Optimization techniques:

The model is improved using TensorFlow's optimization
methods, such as model [5]. enhancing, cutting, and
stemming. The user chooses the display settings from the
regional language or as selected by the user at login.

 Neural Network
A Neural Network is a scientific paradigm that is inspired
by the ways in which organic nervous systems, such as
the brain, process data and pictures into an easily
processed format while retaining the qualities that are
essential for getting superior outcomes.

Fig. 12: Generated image from TensorFlow

"Loss" is the value of the loss over the training data at
the end of each epoch. With training, the optimization
process seeks to decrease this, therefore the lower the
number, the better.

The proportion of accurate predictions to all other

predictions in the training data is referred to as accuracy.
Better if it's higher. Normally, but not always, the loss is
negatively connected with this.

Epoch Train loss Valid loss Time

0 282.509796 284.545929 06:33
1 219.822098 216.609177 06:11
2 184.608017 179.594299 06:16
3 157.201462 152.107513 06:13
Fig. 11: Neural Network 4 146.833130 143.230316 06:08
5 142.532150 140.502411 06:06
Input Layers: In this layer, the model receives input. The 6 130.854355 128.107193 06:13
entire number of characteristics in our data is equal to the 7 123.135384 120.282646 06:16
number of neurons in this layer (number of pixels in case of 8 122.825447 121.192963 06:15
an image). 9 113.097366 110.876205 06:09
 Hidden Layer: The concealed layer is then updated with the 10 110.630867 107.839455 06:05
information from the. Depending on our model and the 11 105.667632 102.160103 06:08
amount of input, hidden layers change. The number of 12 104.367531 101.760544 06:11
neurons in each hidden layer might vary, although they are
13 96.9w82460 92.826294 06:10
often more than the number of features. The output of each
14 94.381241 90.038216 06:11
layer is derived by matrix multiplying the output of the
15 91.442261 87.211021 06:12
layer before it by the learnable weights of that layer, and
then by adding the learnable biases by the activation 16 89.456108 84.471718 06:19
function, which makes the network nonlinear. 17 88.846085 85.127541 06:02
 Output Layer: The output of each class is then converted 18 85.060516 80.282585 06:05
into the probability score for each class using a logistic 19 83.723434 80.030220 06:20
function, such as sigmoid or SoftMax, using the data from Table 1
the hidden layer as input.

IJISRT22JUN602 www.ijisrt.com 815

Volume 7, Issue 6, June – 2022 International Journal of Innovative Science and Research Technology
ISSN No:-2456-2165
V. CONCLUSION

By this Methodology, we can assist with managing all

sorts of identification of problems such as assembly line
product identification, fruits, or vegetables disease
identification. It reduces the number of plants dying due to
improper identification of disease.

REFERENCES

[1.] Savita N. Ghaiwat, Parul Arora, “Detection and

Classification of Plant Leaf Diseases Using Image
processing Techniques: A Review” 3, 2014.
[2.] Renuka Rajendra Kajale, “Detection and
Reorganization of Plant Leaf Disease Using Image
Processing and Android O.S.” March-April 2015.
[3.] A.S.Deokar, Akshay Pophale, Swapnil Patil, Prajakta
Nazarkar, Sukanya Mungase, Harshad Shetye, Tejas
Rane, Tanmay Pawar, Prof. Anuradha Dandwate“ An
Analysis of Methodologies For Leaf Disease Detection
Techniques ” February 2016.
[4.] V Pooja; Rahul Das ; V Kanchana, “Identification of
plant leaf diseases using image processing techniques”,
2017 IEEE Technological Innovations in ICT for
Agriculture and Rural Development (TIAR), April 2018
[5.] Boikobo Tlhobogang ; Muhammad Wannous, “Design
of plant disease detection system: A transfer learning
approach work in progress”, IEEE International
Conference on Applied System Invention (ICASI),
2018,

IJISRT22JUN602 www.ijisrt.com 816