data-science-community-srm / hand-written-digit-classification-recognition Goto Github PK

This project performs digit recognition using deep learning concepts. It can classify an image into 10 classes.We have built a Multilayer perceptron(MLP) using most popular Google library Tensorflow to recognize handwritten digits.

Home Page: https://data-science-community-srm.github.io/Hand-Written-Digit-Classification-Recognition/index.html

License: MIT License

Jupyter Notebook 99.46% HTML 0.34% JavaScript 0.10% CSS 0.10%

python3 deep-learning cnn notebook html-css-javascript

hand-written-digit-classification-recognition's Introduction

ASTRO NUMBERS

This project performs digit recognition using deep learning concepts. It can classify an image into 10 classes.We have built a Multilayer perceptron (MLP) model using most popular Google library Tensorflow to recognize handwritten digits.

A Multi Layer Perceptron is a feedforward neural network, which means that the data is transmitted from the input layer to the output layer in the forward direction.A perceptron is a single neuron model that was a precursor to larger neural networks. The Perceptron consists of an input layer and an output layer which are fully connected.MLPs have the same input and output layers but may have multiple hidden layers in between them.

The MNIST dataset is an acronym that stands for the Modified National Institute of Standards and Technology dataset.It is a dataset of 70,000 small square 28×28 pixel grayscale images of handwritten single digits between 0 and 9.In this project we have used it to classify a given image of a handwritten digit drawn on a canvas into one of 10 classes representing integer values from 0 to 9, inclusively.

Data preprocessing steps:

Splitting the data into training, testing and validation sets.
Flattening the images and displaying it.
Checking the number of instances for each digit.
Plotting graphs and charts for easier understanding.

Model Construction:

Deep learning model: After manually pre-processing the dataset, we come to the part where we use concepts of Multilayer perceptron to build and train a model that classifies the handwritten digits.The 10 classes of digits are 0,1,2,3,4,5,6,7,8,9.

This model used in Astro Numbers has been trained to detect 10 classes of objects: Numbers from 0-9 using deep learning on the MLP model.
It contains 2 hidden layers, 1 dense layer with 512 nodes (ReLU),1 dense layer with 64 nodes (ReLU) and a softmax layer with 10 output nodes using TensorFlow and its libraries.
The model is compiled using the adam optimizer and the categorical cross-entropy loss function will be optimized which is suitable for multi-class classification.We are monitoring the classification accuracy metric since we have the same number of examples in each of the 10 classes.
The final trained model resulted in an accuracy around 97.67% on the dataset with 70,000 images.
The model can be experimented , the user can provide the digit input to the canvas and the model will detect which number it is.

Functionalities

Astro numbers help primary teachers to give the students a unique experience in solving maths problems.
Works on a canvas providing an easy drawing interface.
The web based application interface of Astro numbers uses a MLP model to classify the answers drawn by students.
Assessment can be done by teachers and the final scores will appear on the screen.

Instructions to run:

pip install --upgrade pandas
pip install --upgrade matplotlib
pip install --upgrade seaborn
pip install --upgrade numpy
pip install --upgrade tensorflow
python -m http.server 8000

Project architecture:

Astro-Numbers uses Tensorflow and Keras libraries to build a sequential model with 2 Conv2D layers. We use batch_normalization at the end of every layer for higher accuracy. Activation relu worked best for the dataset. For the output layer, a dense layer was used with softmax activation.The tabular explanation of the same can be seen below.