This document describes a student project to design a handwritten digit recognition system. The main objective is to build a model that can recognize handwritten digits from 0 to 9. The students used the MNIST dataset and explored models like LeNet, ResNet, VGGNet, GoogleNet and CNN. They achieved 98.2% accuracy using LeNet, which had the shortest training time of 7 minutes on average. In conclusion, the students successfully created a digit recognition program using neural networks and datasets.
2. AIM OF THE PROJECT
The main objective of this project is to
design a system which will recognise the
correct handwritten digits written by
humans.
3. PRE-REQUISITE TOOLS AND
LIBRARY
APPLICATIONS
1 2 3
i. Python Programming
ii. Basics of Machine
learning and deep
learning concepts
iii. Tkinter for GUI
i. PyCharm
ii. Jupyter Notebook
iii. Keras Library
iv. MNIST Dataset
v. Tkinter
vi. NumPy
i. Postal Mail Sorting
ii. Bank Cheque
Processing
iii. Form Data Entry
4. STEPS OF PROJECT
DEVELOPMENT
Training the model Evaluating the
model
Parameter tuning
Choosing a model
Preparation of data
Collection of data
1 2 3
4 5 6 7
Making predictions
6. MNIST Dataset
Among thousands of datasets available in the market, MNIST is the most popular
dataset for enthusiasts of machine learning and deep learning.
Above 60,000 plus training images of handwritten digits from 0 to 9 and more
than 10,000 images for testing are present in the MNIST dataset. MNIST dataset
can be directly in- built in keras, we don’t have to download the dataset.
8. VGGNet
VGG stands for Visual Geometry Group; it is a standard deep
Convolutional Neural Network (CNN) architecture with multiple
layers. The “deep” refers to the number of layers.
Average Training Time: 15-30 mins
Accuracy: 99%
9. LeNet
It is a simple convolutional neural network that held the basis for
subsequent advancements in deep learning and image recognition.
Average Training Time: 5-10 mins
Accuracy: 98-99%
10. ResNet
ResNet, short for Residual Network, is a deep convolutional neural network
(CNN) architecture. ResNet architectures typically have hundreds or even
thousands of layers, enabling them to learn complex patterns and achieve state-
of-the-art results on various computer vision tasks, including image classification,
object detection, and image segmentation.
Average Training Time: 10-20 mins
Accuracy: 99.3-99.5%
11. GoogleNet
GoogLeNet, also known as Inception v1, is a deep convolutional
neural network (CNN) architecture developed by researchers at
Google. GoogLeNet, or Inception v1, has 22 layers in total. It includes
both convolutional layers and fully connected layers.
Average Training Time: Few hours to one day
Accuracy: 98-99%
12. CNN
Convolutional Neural Network is a type of artificial neural network
specifically designed for processing and analyzing structured grid-like
data, such as images or sequences.
Average Training Time: Few hours to one day
Accuracy: 98-99%
15. RESULT
After learning and working with all the models we decided to work with the LeNet-5
model for our project and successfully build a digit recognition system with 98.2%
accuracy and an average training time of 7.1 minutes.
ResNet CNN
GoogLeNet
VGGNet
LeNet
16. CONCLUSION
We have successfully completed our project and built a program to determine
handwritten digits successfully using various datasets and neural networks.
19. VGGNet
Briefly elaborate on what you want to discuss.
Add a main poit Briefly elaborate on what you want to discuss.
Add a main point Briefly elaborate on what you want to discuss.
Add a main point Briefly elaborate on what you want to discuss.
Add a main point Briefly elaborate on what you want to discuss.
Add a main point Briefly elaborate on what you want to discuss.
20. ST QUARTER
January
• Double click to add task
February
• Double click to add task
March
• Double click to add task
2ND QUARTER
April
• Double click to add task
May
• Double click to add task
June
• Double click to add task
3RD QUARTER
July
• Double click to add task
August
• Double click to add task
September
• Double click to add task
4TH QUARTER
October
• Double click to add task
November
• Double click to add task
December
• Double click to add task
Comparing all 5 neural networks
21. how to use this
presentation
• Click on the "Google Slides"
button below this presentation
preview.
• Click on "Make a copy."
• Start editing your presentation.
• You need to sign in to your
Google account.
Google Slides
• Click on the "PowerPoint" button
below this presentation preview.
• Start editing your presentation.
• Download and install the fonts
used in this presentation as
listed on the next page.
PowerPoint
• Click on the "Canva" button
under this presentation preview.
• Start editing your presentation.
• You need to sign in to your
Canva account.
Canva
22. download this presentation as
a Google Slides template in 3 steps
Click on the "Canva" button
next to this presentation. Start
editing your presentation. You
may need to sign in to your
Canva account.
Click on the "Share" button on
the top right corner in Canva
and click on "More" at the
bottom of the menu.
Enter "Google Drive" in the search
bar and click on the Google Drive
icon to add this design as a Google
slide in your G Drive. It will ask you
to link your Google account to
Canva the first time.
1 2 3
26. Elaborate on what you want to discuss.
study
objectives
add a main point
Elaborate on what you want to discuss.
add a main point
Elaborate on what you want to discuss.
add a main point
27. add a main
point
Elaborate on what
you want to
discuss.
methodology
add a main
point
Elaborate on what
you want to
discuss.
add a main
point
Elaborate on what
you want to
discuss.
28. Elaborate on the featured
statistic.
2 out of 5 95% 12 million
Elaborate on the featured
statistic.
Elaborate on the featured
statistic.
33. add a team
members page
Elaborate on what you want to discuss.
Name
Title or Position
Name
Title or Position
Name
Title or Position
Name
Title or Position
35. fonts in this
presentation
This presentation template uses the
following free fonts:
• Titles: Playfair Display Black
• Headers: Roboto
• Body Copy: Roboto
You can find these fonts online too.
Happy designing!
36. swot analysis
W
O
T
What are you doing well?
What sets you apart?
What are your good qualities?
Where do you need
to improve?
Are resources adequate?
What do others do better
than you?
What are your goals?
Are demands shifting?
How can it be improved?
What are the blockers
you're facing?
What are factors outside of
your control?
37. Item Item Item Item Item
25
20
15
10
5
0
analysis of the
results
Elaborate on what you want
to discuss.
39. resource
page
Use these design resources in your
Canva Presentation. Happy
designing!
Don't forget to delete this page
before presenting.
40. free templates for all your
presentation needs
For PowerPoint,
Google Slides, and
Canva
100% free for
personal or
commercial use
Ready to use,
professional, and
customizable
Blow your audience
away with attractive
visuals