표 3. | Table 3. 제안된 방식, DopplerNet, pretrained 모델의 F1 score 비교 | F1 score comparison between proposed method, DopplerNet, and pretrained models.

Model type	Input timestep	Integration type	F1 score	Neural network parameter
LeNet[15]	1	-	0.9641	5,602,979
	2	NCI	0.9749	5,602,979
	3	NCI	0.9840	5,602,979
ResNet18	1	-	0.9571	11,178,051
	2	NCI	0.9749	11,178,051
	3	NCI	0.9825	11,178,051
ResNet50	1	-	0.9592	23,514,179
	2	NCI	0.9691	23,514,179
	3	NCI	0.9798	23,514,179
VGG19[16]	1	-	0.9559	139,593,539
	2	NCI	0.9746	139,593,539
	3	NCI	0.9813	139,593,539
NasNetMobile[1],[17]	3	Channel-wise concatenation	0.9769	4,272,887
MobileNetV2[1],[18]	3	Channel-wise concatenation	0.9894	3,325,043
DopplerNet[1]	3	Channel-wise concatenation	0.9948	3,818,755
Proposed method	1	-	0.9727	1,406,473
	2	NCI	0.9850	1,406,473
	3	NCI	0.9912	1,406,473