Handwritten-Equation-Recognition---CNN

A WebApp for Solve Handwritten Equations using Convolutional Neural Networks

It works with any digit from 0-9 and the symbols +,-,x,√

White background and black digits are necessary

Convolutional Neural Network

• Two Convolutional Layers
• Two MaxPooling Layers
• Dropout Layer
• Flatten Layer
• Two Dense Layers using Relu
• Output Layer using Softmax
• Adam Optimizer

Installation

1. Create a virtual environment:

python3 -m venv venv

2. Activate the virtual environment:

   • On Linux

    . venv/bin/activate

* On Windows

    venv\Scripts\activate

3. Install the dependencies

	pip install -r requirements.txt

4. If port if already in use

    pkill -9 python

Usage

In the folder model, there is a pre-build model of the neural network and the weights of that model. But if you wanna build or train in your own hands, you need to create a folder called 'datasets' in the root of the project, download the datasets from here and extract the folders 0, 1, 2, 3, 4 ,5, 6, 7, 8, 9, +, -, times (rename the folder for x)', sqrt (rename the folder for √) in the folder

After that you are ready to start the application. For that is just:

python -m flask run --without-threads

Used Technologies

• TensorFlow / Keras
• Python / Flask
• OpenCv
• P5.js