. 2020 Sep 17;20(18):5315. doi: 10.3390/s20185315

Table 5.

DenseNet Network Structure in the Improved SSD Algorithm for Solid Wood Board Defect Recognition.

Layers	DenseNet121 (k = 32)	Output Size	Channels
Input		200 × 200	3
Convolution	7 × 7 conv, stride 2	100 × 100	64
Pooling	3 × 3 max pool, stride 2	50 × 50	64
Dense Block (1)	Bottleneck × 6	50 × 50	256
Transition Layer (1)	1 × 1 conv	50 × 50	128
Transition Layer (1)	2 × 2 average pool, stride 2	25 × 25	128
Dense Block (2)	Bottleneck × 12	25 × 25	512
Transition Layer (2)	1 × 1 conv	25 × 25	256
Transition Layer (2)	2 × 2 average pool, stride 2	12 × 12	256
Dense Block (3)	Bottleneck × 24	12 × 12	1024
Transition Layer (3)	1 × 1 conv	12 × 12	512
Transition Layer (3)	2 × 2 average pool, stride 2	6 × 6	512
Dense Block (4)	Bottleneck × 16	6 × 6	1024
Conv Layer	1 × 1 conv	6 × 6	1024