Alphabet Recognition System using Convolutional Neural Network (CNN)

Convolutional Neural Network (CNN) is a Deep Learning Algorithm widely used for character recognition.This algorithm is integrated with anvil website which identifies the alphabet present in the given input image.

File Description

AlphabetRecognitionfinal.ipynb - This jupyter notebook contains the algorithm for the CNN model.

Training.zip and Testing.zip - These folders contain images of alphabets ranging from a to z.

CNN_model.sav - This file contains the trained CNN model for identifying alphabets from images. This is a ready to use model and can directly be loaded to test images using the following command-

  import pickle
  model = pickle.load(open('CNN_model.sav','rb'))

Performance

Dataset

Training Dataset - 501 images belonging to 26 classes (a-z)

Testing Dataset - 260 images belonging to 26 classes (a-z)

Model


  Model: "sequential_7"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
conv2d_13 (Conv2D)           (None, 30, 30, 32)        896       
_________________________________________________________________
max_pooling2d_13 (MaxPooling (None, 15, 15, 32)        0         
_________________________________________________________________
conv2d_14 (Conv2D)           (None, 13, 13, 32)        9248      
_________________________________________________________________
max_pooling2d_14 (MaxPooling (None, 6, 6, 32)          0         
_________________________________________________________________
flatten_7 (Flatten)          (None, 1152)              0         
_________________________________________________________________
dense_13 (Dense)             (None, 128)               147584    
_________________________________________________________________
dense_14 (Dense)             (None, 26)                3354      
=================================================================
Total params: 161,082
Trainable params: 161,082
Non-trainable params: 0
_________________________________________________________________

Accuracy

  
Epoch 1/3
16/16 [==============================] - 1s 62ms/step - loss: 0.1866 - accuracy: 0.9082 - val_loss: 0.7112 - val_accuracy: 0.8657
Epoch 2/3
16/16 [==============================] - 1s 60ms/step - loss: 0.1769 - accuracy: 0.9301 - val_loss: 0.4118 - val_accuracy: 0.8662
Epoch 3/3
16/16 [==============================] - 1s 54ms/step - loss: 0.1672 - accuracy: 0.9261 - val_loss: 0.2001 - val_accuracy: 0.9342

Check out my medium article for a step by step tutorial on building a CNN model for Alphabet Recognition and deploying it with Anvil.

https://medium.com/@sakshibutala12/building-and-deploying-an-alphabet-recognition-system-7ab59654c676?sk=b6f75f1639e8c301995b412b74d589ca

Author

Sakshi Butala