An Action Recognition Method for Volleyball Players Using Deep Learning

Table 4

Influence of different convolution kernel sizes on the recognition effect.

	Combination 1	Combination 2	Combination 3	Combination 4
Convolution kernel size Average recognition rate	C1: 9 × 9	C1: 11 × 11	C1: 13 × 13	C1: 15 × 15
	C2: 7 × 7	C2: 7 × 7	C2: 8 × 8	C2: 8 × 8
Florence3D action (%)	91.00	92.00	93.00	92.00

Note. C1 is the first convolution, and C2 is the second convolution.