Research on Classification Method of Maize Seed Defect Based on Machine Vision

Table 2

The network structure based on VGG19, a global average pooling layer, and two dense layers were added for transfer learning. Params label means the number of parameters per layer.

No.	Layer name	Output shape	Params
1	input_1(InputLayer)	(224,224,3)	0
2	block1_conv1(Conv2D)	(224, 224, 64)	1792
3	block1_conv2 (Conv2D)	(224, 224, 64)	36928
4	block1_pool (MaxPooling2D)	(112, 112, 64)	0
5	block2_conv1 (Conv2D)	(112, 112, 128)	73856
6	block2_conv2 (Conv2D)	(112, 112, 128)	147584
7	block2_pool (MaxPooling2D)	(56, 56, 128)	0
8	block3_conv1 (Conv2D)	(56, 56, 128)	295168
9	block3_conv2(Conv2D)	(56, 56, 256)	590080
10	block3_conv3 (Conv2D)	(56, 56, 256)	590080
11	block3_conv4(Conv2D)	(56, 56, 256)	590080
12	block3_pool (MaxPooling2D)	(28, 28, 256)	0
13	block4_conv1(Conv2D)	(28, 28, 512)	1180160
14	block4_conv2 (Conv2D)	(28, 28, 512)	2359808
15	block4_conv3(Conv2D)	(28, 28, 512)	2359808
16	block4_conv4 (Conv2D)	(28, 28, 512)	2359808
17	block4_pool(MaxPooling2D)	(14, 14, 512)	0
18	block5_conv1 (Conv2D)	(14, 14, 512)	2359808
19	block5_conv2(Conv2D)	(14, 14, 512)	2359808
20	block5_conv3 (Conv2D)	(14, 14, 512)	2359808
21	block5_conv4(Conv2D)	(14, 14, 512)	2359808
22	MaxPooling2D	(7, 7, 512)	0
23	global_average_pooling2d	(512)	0
24	dense (Dense)	(1024)	525312
25	dense_1 (Dense)	(2)	2050