Multifeature Fusion Human Motion Behavior Recognition Algorithm Using Deep Reinforcement Learning

Table 2

Parameter settings of deep reinforcement learning network recognition model.

No.	Parameter item	Value
1	Number of hidden units	512
2	Sequence length value	10
3	Model initial parameter acquisition method	CNN
4	Minimum group size	32
5	Momentum factor	0.9
6	Weight attenuation	5 × 10^−4
7	Initial learning rate	10^–3
8	Number of training rounds	128
9	Select frame policy superparameters	10
10	Select image block policy superparameters	70
11	Frame relative position	1–10 frames
12	Frame sequence length	10
13	Image block threshold	70 pixels