Fingerspelling Identification for Chinese Sign Language via AlexNet-Based Transfer Learning and Adam Optimizer

Table 1

Learnable layers of AlexNet.

Name	Weights	Biases
CL1	11 ∗ 11 ∗ 3 ∗ 96 = 34,848	1 ∗ 1 ∗ 96 = 96
CL2	5 ∗ 5 ∗ 48 ∗ 256 = 307,200	1 ∗ 1 ∗ 256 = 256
CL3	3 ∗ 3 ∗ 256 ∗ 384 = 884,736	1 ∗ 1 ∗ 384 = 384
CL4	3 ∗ 3 ∗ 192 ∗ 384 = 663,552	1 ∗ 1 ∗ 384 = 384
CL5	3 ∗ 3 ∗ 192 ∗ 256 = 442,368	1 ∗ 1 ∗ 256 = 256
FCL6	4096 ∗ 9216 = 37,748,736	4096 ∗ 1 = 4,096
FCL7	4096 ∗ 4096 = 16,777,216	4096 ∗ 1 = 4,096
FCL8	1000 ∗ 4096 = 4,096,000	1000 ∗ 1 = 1,000
Total	60,954,656	10,568