The Research of Chronic Gastritis Diagnosis with Electronic Noses

Table 1

Description of proposed twenty layer DCNN architecture.

Layers	Layer name	Maps and neurons	Kernel	params	ops
0	Image input	3@56x56	—
1	Convolution	96@14x14	11x11	12 K	2.3 M
2	Batch normalization	Batch normalization	—
3	ReLu	ReLu	—
4	Max-pooling	96@7x7	3x3
5	Convolution	256@7x7	5x5	6 K	0.3 M
6	Batch normalization	Batch normalization	—
7	ReLu	ReLu	—
8	Max-pooling	256@4x4	3x3
9	Convolution	384@4x4	3x3	4 K	0.06 M
12	Batch normalization	Batch normalization	—
13	ReLu	ReLu	—
14	Max-pooling	384@2x2	3x3
16	Fully connected	256	—	393 K	0.4 M
17	Dropout	Dropout	—
18	Fully connected	256	—	65 K	0.07 M
19	Dropout	Dropout	—
20	Classification output	2	—	512	512