Table 1.

DCNN structure properties before and after pruning.

Layer	Num. of Neurons	Num. of Nodes (filter size)	Num. of Parameters
	Before pruning
C1	61,504	64 (11 × 11)	23,296
C2	43,200	192 (5 × 5)	307,392
C3	18,816	384 (3 × 3)	663,936
C4	12,544	256 (3 × 3)	884,992
C5	12,544	256 (3 × 3)	590,080
F1		4096	9,441,280
F2		4096	16,781,312
F3		1000	4,097,000
F4		100	100,100
F5		2	202
Total	148,608	Convolutional: 1,152 Fully-connected: 9,294	32,889,590
	After pruning
C1	15,376	16 (11 × 11)	5,824
C2	1,350	16 (5 × 5)	6,416
C3	786	16 (3 × 3)	2,320
C4	786	16 (3 × 3)	2,320
C5	786	16 (3 × 3)	2,320
F1		4096	593,920
F2		4096	16,781,312
F3		1000	4,097,000
F4		100	100,100
F5		2	202
Total	19,084	Convolutional: 80 Fully connected: 9,294	21,591,734
Decrease in number of neurons: 87.2%			Decrease in number of parameters: 34.4%