A Video Based Fire Smoke Detection Using Robust AdaBoost

. 2018 Nov 5;18(11):3780. doi: 10.3390/s18113780

Algorithm 1 Improved Robust AdaBoost classifier with RBFSVM: RAB-RBFSVM+.

Input: Training datasets

s = {\{(x_{i}, y_{i})\}}_{i = 1}^{m}

with labels

y \in \{- 1, + 1\}

1:
initialize: Normalize training datasets; Initialize weights of datasets via Equation (46), $σ_{s t e p}$ via Equation (41); $σ_{i n i} \leftarrow 12$ ; $m \leftarrow 1; e r r \leftarrow - 1; N \leftarrow 100; I_{0_t r} \leftarrow I$ ;
2:
while $m < N & e r r \neq 0$ do
3:
Train a RBFSVM base classifier ht on the weighted training datasets and get the misclassification flag of samples $I_{m}$ .
4:
Compute training errors using Equation (28)
5:
Compute diversity value using Equations (39) and (40)
6:
Compute misclassification times of samples: $I_{m_t r} = I_{m - 1_t r} + I_{m}$
7:
Compute possibility of noise samples via Equations (42)–(44)
8:
Decrease $σ$ by $σ_{m}$
9:
if $σ > σ_{s t e p} & e r r \neq 0$ then
10:
Catch the right $σ$ which minimize the error: $σ_{m}$ , and go to step 3
11:
end if
12:
$σ = σ_{m}$
13:
Repeat step 3-7
14:
Set the weight of component classifier via Equation (34)
15:
Update the weight of samples via Equation (45)
16:
end while

Output: Combine classifiers via Equations (32) and (33)