This model involves recognition of hindi hand written characters using Convolutional neural network. Python implementation using keras has been done. The dataset can be downloaded from the following link: https://archive.ics.uci.edu/ml/datasets/Devanagari+Handwritten+Character+Dataset
The dataset contains 92,000 images of handwritten hindi characters belonging to 46 classes. The data is splitted into training set(85%) and test set(15%). The images are of size 32x32 in .png format.
The architecture involves four CNN layers followed by 3 fully connected layer. The final layers using softmax function helps to provide the most probable answer.
Loss function- Categorical Cross entropy Optimizer- Adam Final output layer activation- Softmax
Model Train file- Hindi.py Testing individual images- check.py
Accuracy achieved in 25 epochs- 98.94%