Benchmark Pashto Handwritten Character Dataset and Pashto Object Character Recognition (OCR) Using Deep Neural Network with Rule Activation Function

Table 3

Parameters’ details of used models.

	Model 1	Model 2	Model 3
Learning rate	0.0001	0.0001	0.0001
Algorithm	Backpropagation with Softmax on the last layer	Backpropagation with Softmax on the last layer	Backpropagation with Softmax on the last layer
ReLU activation layers	Hidden layer 1	Hidden layer 1, Hidden Layer 2	Hidden layer 1, Hidden Layer 2, Hidden Layer 3
Loss function	Cross-entropy	Cross-entropy	Cross-entropy
Epochs	10000	10000	10000
Dataset size	43000 samples	43000 samples	43000 samples
Training data ratio	90% (38700 sample images)	90% (38700 sample images)	90% (38700 sample images)
Testing data ratio	10% (4300 sample images)	10% (4300 sample images)	10% (4300 sample images)
Tensor flow function	Gradient descent with Softmax	Gradient descent with Softmax	Gradient descent with Softmax