Smoke Detection in Storage Yard Based on Parallel Deep Residual Network

Zhenglai Wang; Min Huang; Qibing Zhu; Sheng Jiang

doi:10.3788/LOP55.051008

Journals >Laser & Optoelectronics Progress >Volume 55 >Issue 5 >Page 051008 > Article

Laser & Optoelectronics Progress
Vol. 55, Issue 5, 051008 (2018)

Smoke Detection in Storage Yard Based on Parallel Deep Residual Network

Zhenglai Wang, Min Huang^*, Qibing Zhu, and Sheng Jiang

Author Affiliations

Key Laboratory of Advanced Process Control for Light Industry, Ministry of Education, Jiangnan University, Wuxi, Jiangsu 214122, China

show less

DOI: 10.3788/LOP55.051008 Cite this Article Set citation alerts

Zhenglai Wang, Min Huang, Qibing Zhu, Sheng Jiang. Smoke Detection in Storage Yard Based on Parallel Deep Residual Network[J]. Laser & Optoelectronics Progress, 2018, 55(5): 051008 Copy Citation Text

show less

Fig. 1. Overall framework of the method in this study

Download full size

Fig. 2. Schematic of the like-smoke object

Download full size

Fig. 3. Residual block

Download full size

Fig. 4. Residual network

Download full size

Fig. 5. Part of the experimental videos

Download full size

Fig. 6. Part of the test results

Download full size

Fig. 7. ROC curves of validation scenarios with parallel and single network

Download full size

Fig. 8. ROC curves of test scenarios with parallel and single network

Download full size

Layer name	Output size	50-layer
conv1	112×112	74×74, 64, stride2
conv2_x	56×56	3×3 max pool, stride2
		$\begin{matrix} [\begin{matrix} 1 \times 1,64 \\ 3 \times 3,64 \\ 1 \times 1,256 \end{matrix}] \end{matrix}$ ×2, stride1
		$\begin{matrix} [\begin{matrix} 1 \times 1,64 \\ 3 \times 3,64 \\ 1 \times 1,256 \end{matrix}] \end{matrix}$ ×1, stride2
conv3_x	28×28	$\begin{matrix} [\begin{matrix} 1 \times 1,128 \\ 3 \times 3,128 \\ 1 \times 1,512 \end{matrix}] \end{matrix}$ ×3, stride1
conv3_x	28×28	$\begin{matrix} [\begin{matrix} 1 \times 1,128 \\ 3 \times 3,128 \\ 1 \times 1,512 \end{matrix}] \end{matrix}$ ×1, stride2
conv4_x	14×14	$\begin{matrix} [\begin{matrix} 1 \times 1,256 \\ 3 \times 3,256 \\ 1 \times 1,1024 \end{matrix}] \end{matrix}$ ×5, stride1
conv4_x	14×14	$\begin{matrix} [\begin{matrix} 1 \times 1,256 \\ 3 \times 3,256 \\ 1 \times 1,1024 \end{matrix}] \end{matrix}$ ×1, stride2
conv5_x	7×7	$\begin{matrix} [\begin{matrix} 1 \times 1,512 \\ 3 \times 3,512 \\ 1 \times 1, 2048 \end{matrix}] \end{matrix}$ ×3, stride1
	1×1	Average pool,1000-d fc,softmax

Table 1. Network structure of ResNet 50

Confidencelevel	Single network		Parallel network
Confidencelevel	Detection rate	False positive rate	Detection rate	False positive rate
0.1	93.764	1.357	95.768	0.581
0.2	92.650	0.678	94.655	0.194
0.3	91.759	0.581	92.984	0.097
0.4	90.646	0.484	91.314	0.000
0.5	89.978	0.291	90.535	0.000
0.6	88.864	0.097	89.421	0.000
0.7	87.751	0.097	88.085	0.000
0.8	85.412	0.000	85.746	0.000
0.9	81.849	0.000	81.849	0.000
0.95	78.953	0.000	79.065	0.000
0.98	73.385	0.000	73.608	0.000

Table 2. Detection rate and false positive rate of validation scenarios with single and parallel network%

Confidence level	Single network		Parallel network
Confidence level	Detection rate	False positive rate	Detection rate	False positive rate
0.1	43.363	26.540	69.912	36.967
0.2	37.168	15.640	65.487	16.588
0.3	34.513	15.640	61.504	13.744
0.4	31.858	15.640	58.850	12.322
0.5	30.973	15.166	56.637	11.848
0.6	29.204	13.744	54.867	10.427
0.7	27.876	12.322	52.212	8.531
0.8	26.549	9.953	49.115	7.109
0.9	24.779	0.474	45.575	0.000
0.95	19.027	0.000	38.496	0.000
0.98	12.389	0.000	26.549	0.000

Table 3. Detection rate and false positive rate of test scenarios with single and parallel network%

Zhenglai Wang, Min Huang, Qibing Zhu, Sheng Jiang. Smoke Detection in Storage Yard Based on Parallel Deep Residual Network[J]. Laser & Optoelectronics Progress, 2018, 55(5): 051008

Download Citation

Set citation alerts for the article

Tools

Set citation alerts for the article

Save the article for my favorites

Paper Information