Enhancing the Power of CNN Using Data Augmentation Techniques for Odia Handwritten Character Recognition

Table 1

Architectural specification of the proposed five CNN models M1, M2, M3, M4, and M5.

Layer		Feature map					Size					Kernel size	Stride	Activation
		Model no.					Model no.
		1	2	3	4	5	1	2	3	4	5
Input	Image	1	1	1	1	1	28 × 28					—	—	—
1	Convolution	45	45	25	5	20	24 × 24					5	1	ReLu
2	Max pooling	45	45	25	5	20	12 × 12					2	2	—
3	Batch normalization	45	45	25	5	20	12 × 12					2	-	—
4	Convolution	75	50	75	75	70	8 × 8					5	1	ReLu
5	Max pooling	75	50	75	75	70	4 × 4					2	2	—
6	Batch normalization	75	50	75	75	70	4 × 4					2	—	—
7	Flatten	—	—	—	—	—	1200	800	1200	1200	1120	—	—	—
8	Dense	—	—	—	—	—	300	200	300	300	300	—	—	ReLu
9	Batch normalization	—	—	—	—	—	300	200	300	300	300	—	—	—
Output		—	—	—	—	—	47	47	47	47	47	—	—	Softmax