Region Anomaly Detection via Spatial and Semantic Attributed Graph in Human Monitoring

. 2023 Jan 23;23(3):1307. doi: 10.3390/s23031307

Algorithm 1 Overall procedure of SSGAE.

Input: Graph

G_{train}^{k} = {A^{k}, X^{k}}_{k = 1}^{K}

G_{test}^{k^{'}} = {A^{k^{'}}, X^{k^{'}}}_{k^{'} = 1}^{K^{'}}

; Learnable parameter

Θ

; Hyper-parameter

β

; Number L of the hidden layers in SSGAE; Number T of the training epochs.
Output: Anomaly score

s_{i}^{k^{'}}

for each node

v_{i}^{k^{'}}

via function

f (\cdot)

1:
▹ Training Stage.
2:
Randomly initialize $Θ$ and the trainable parameters in ${MLP}_{Enc}$ , ${MLP}_{Str - Dec}$ and ${MLP}_{Att - Dec}$ ;
3:
for $t = 1, 2, \dots, T$ do;
4:
for $k = 1, 2, \dots, K$ do
5:
for $l = 1, 2, \dots, L$ do
6:
Calculate $H^{(l)}$ via Equation (3);
7:
end for
8:
$Z^{k} = H^{(L)}$ ;
9:
for $l = 1, 2, \dots, L$ do
10:
Calculate ${\hat{H}}^{(l)}$ via Equation (6);
11:
end for
12:
${\hat{X}}^{k} = {\hat{H}}^{(L)}$ ;
13:
Calculate ${\hat{A}}^{k}$ via Equation (4);
14:
Update $Θ$ and the trainable parameters in ${MLP}_{Enc}$ , ${MLP}_{Str - Dec}$ , and ${MLP}_{Att - Dec}$ via Equation (8) with the backpropagation algorithm.
15:
end for
16:
end for
17:
▹ Test Stage.
18:
for $k^{'} = 1, 2, \dots, K^{'}$ do
19:
for $l = 1, 2, \dots, L$ do
20:
Calculate $H^{(l)}$ via Equation (3);
21:
end for
22:
$Z^{k^{'}} = H^{(L)}$ ;
23:
for $l = 1, 2, \dots, L$ do
24:
Calculate ${\hat{H}}^{(l)}$ via Equation (6);
25:
end for
26:
${\hat{X}}^{k^{'}} = {\hat{H}}^{(L)}$ ;
27:
Calculate ${\hat{A}}^{k^{'}}$ via Equation (4);
28:
Calculate anomaly score $s_{i}^{k^{'}}$ of each node $v_{i}^{k^{'}}$ in $G_{test}^{k^{'}}$ via Equation (9).
29:
end for