Researching | Video smoke recognition based on random patch shift and deformable attention

Journal of Applied Optics, Volume. 45, Issue 6, 1204(2024)

Video smoke recognition based on random patch shift and deformable attention

Yehui XIE and Haitao ZHAO^*

Author Affiliations

School of Information Science and Engineering, East China University of Science and Technology, Shanghai 200237, China

show less

Abstract Get PDF(in Chinese)

Figures&Tables (12)

References (18)

Paper Information

Figures & Tables(12)

Fig. 1. Overall framework diagram of network

Download full size

View in Article

Fig. 2. Different patch shift patterns

Download full size

View in Article

Fig. 3. Schematic diagram of deformable attention

Download full size

View in Article

Fig. 4. Deformable attention

Download full size

View in Article

Fig. 5. Overview of datasets

Download full size

View in Article

Fig. 6. Grad-CAM visualization

Download full size

View in Article

Table 1. Comparison of different methods on RISE test set

View in Article

Table 1. Comparison of different methods on RISE test set

方法	$F_{1}$ 分数
方法	$S_{0}$	$S_{1}$	$S_{2}$	$S_{3}$	$S_{4}$	$S_{5}$
Flow-SVM	0.42	0.59	0.47	0.63	0.52	0.47
Flow-I3D	0.55	0.58	0.51	0.68	0.65	0.50
SVM	0.57	0.70	0.67	0.67	0.57	0.53
I3D	0.80	0.84	0.82	0.87	0.82	0.75
I3D-ND	0.76	0.79	0.81	0.86	0.76	0.68
I3D-FP	0.76	0.81	0.82	0.87	0.81	0.71
I3D-TSM	0.81	0.84	0.82	0.87	0.80	0.74
I3D-LSTM	0.80	0.84	0.82	0.85	0.83	0.74
I3D-TC	0.81	0.84	0.84	0.87	0.81	0.77
CNN-NonFFM^[12]	0.83	0.82	0.84	0.85	0.78	0.83
EFFNet^[12]	0.84	0.83	0.86	0.86	0.80	0.83
AFSNet^[13]	0.85	0.86	0.82	0.91	0.81	0.80
本文方法	0.85	0.85	0.86	0.88	0.84	0.79

Table 2. Performance comparison of different methods
View table
View in Article
Table 2. Performance comparison of different methods
方法 Parameters/M FLOPs/G FPS
I3D 12.3 62.7 32.71
I3D-TSM 12.3 62.7 31.40
I3D-LSTM 38.0 62.9 32.25
I3D-TC 12.3 62.7 32.88
EFFNet^[12] 27.2 34.6 42.57
AFSNet^[13] 30.8 40.6 34.87
本文方法 24.2 68.4 31.78

Table 3. Ablation experiments of RPS and DA
View table
View in Article
Table 3. Ablation experiments of RPS and DA
模型 $F_{1}$ 分数 $A_{cc}$ $P_{r}$ $R_{e}$
Swin 0.580 2 0.704 9 0.615 7 0.548 6
Swin+RPS 0.846 5 0.886 9 0.853 8 0.839 4
Swin+RPS+DA 0.850 8 0.892 5 0.879 2 0.824 2

Table 4. Ablation experiments of RPS in different layers of Swin Transformer
View table
View in Article
Table 4. Ablation experiments of RPS in different layers of Swin Transformer
层 $F_{1}$ 分数 $A_{cc}$ $P_{r}$ $R_{e}$
1 2 3 4
$\sqrt$ 0.808 9 0.872 7 0.915 6 0.724 4
$\sqrt$ $\sqrt$ 0.829 9 0.880 9 0.884 3 0.781 9
$\sqrt$ $\sqrt$ $\sqrt$ 0.845 8 0.888 1 0.866 9 0.825 9
$\sqrt$ $\sqrt$ $\sqrt$ $\sqrt$ 0.846 5 0.886 9 0.853 8 0.839 4

Table 5. Different patterns of RPS
View table
View in Article
Table 5. Different patterns of RPS
模式 $F_{1}$ 分数 $A_{cc}$ $P_{r}$ $R_{e}$
pattern-a 0.838 7 0.883 1 0.861 1 0.817 4
pattern-b 0.834 7 0.883 1 0.880 0 0.793 7
pattern-c 0.850 8 0.892 5 0.879 2 0.824 2

Table 6. Ablation experiments of DA in different layers of Swin Transformer+RPS
View table
View in Article
Table 6. Ablation experiments of DA in different layers of Swin Transformer+RPS
层 $F_{1}$ 分数 $A_{cc}$ $P_{r}$ $R_{e}$
3 4
$\sqrt$ 0.850 8 0.892 5 0.879 2 0.824 2
$\sqrt$ $\sqrt$ 0.809 2 0.866 4 0.863 0 0.761 6

Tools

Get Citation

Copy Citation Text

Yehui XIE, Haitao ZHAO. Video smoke recognition based on random patch shift and deformable attention[J]. Journal of Applied Optics, 2024, 45(6): 1204

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category:

Received: Sep. 21, 2023

Accepted: --

Published Online: Jan. 14, 2025

The Author Email: ZHAO Haitao (赵海涛)

DOI:10.5768/JAO202445.0602005

Topics

laser devices and laser physics

Lasers and Laser Optics

laser manufacturing

Instrumentation, Measurement and Metrology