Research on Fire Detection Method based on Deep Neural Network MODIS Data

火灾名称	火灾范围	纬度	经度	发现日期	扑灭日期
INOWAK	606 945	61.982	157.086	6/25/1997	9/9/1997
LONG DRAW	558 198.3	42.391 89	117.894	7/8/2012	7/30/2012
WALLOW	538 049	33.606 11	109.45	5/29/2011	7/12/2011
BOUNDARY	537 627	65.266 3	146.886	6/13/2004	9/30/2004

Table 2. Table 2Characteristics of the MOD021KM/MYD021KM

View table

View in Article

Table 2. Table 2Characteristics of the MOD021KM/MYD021KM

通道	波长分布/nm	信噪比	空间分辨率/m
1	620~670	128	1 000
2	841~876	201	1 000
3	459~479	243	1 000
4	545~565	228	1 000
5	1 230~1 250	74	1 000
6	1 628~1 652	275	1 000
7	2 105~2 155	110	1 000
21	3 929~3 989	2.00	1 000
22	3 929~3 989	0.07	1 000
31	1 078~1 128	0.05	1 000
32	1 177~1 227	0.05	1 000

Table 3. Table 3Description of Brightness Temperature Mean and Mean Absolute Deviation

View table

View in Article

Table 3. Table 3Description of Brightness Temperature Mean and Mean Absolute Deviation

计算公式	含义
$Δ B T_{4} = \frac{\sum_{i = 1}^{n} (B T_{4} - B T_{4 i})}{n}$	4 um的中心像元与背景像元的亮度温度差
$Δ B T_{11} = \frac{\sum_{i = 1}^{n} (B T_{11} - B T_{11 i})}{n}$	11 um的中心像元与背景像元的亮度温度差
$Δ B T_{b 4 - b 11} = Δ B T_{4} - Δ B T_{11}$	中心与背景像元在4 um与11 um亮温的差值平均值
$δ_{4}$ = $\frac{\sum_{i = 1}^{n} \|B T_{4 i} - \frac{\sum_{i = 1}^{n} B T_{4 i}}{n}\|}{n}$	4 um背景窗口亮度温度平均绝对偏差
$δ_{11}$ = $\frac{\sum_{i = 1}^{n} \|B T_{11 i} - \frac{\sum_{i = 1}^{n} B T_{11 i}}{n}\|}{n}$	11 um背景窗口亮度温度平均绝对偏差
${δ_{b 4}}_{- b 11}$ = $\frac{\sum_{i = 1}^{n} \|(B T_{4 i} - B T_{11 i}) - \frac{\sum_{i = 1}^{n} (B T_{4 i} - B T_{11 i})}{n}\|}{n}$	4 um与11 um背景亮度温差平均绝对偏差

Table 4. Table 4Network Input Characteristics
View table
View in Article
Table 4. Table 4Network Input Characteristics
热红外特征反射率特征指数特征
$B T_{4}$
$ρ_{1}$
$ρ_{2}$
$ρ_{5}$
$ρ_{7}$
$N D V I$
$N D I I 7$
$B T_{11}$
$B T_{12}$
$B T_{4} - 11$
$Δ B T_{4}$
$Δ B T_{11}$
$Δ B T_{b 4 - b 11}$
$δ_{4}$
$δ_{11}$
${δ_{b 4}}_{- b 11}$

Table 5. Table 5Fire test set results for different input configurations
View table
View in Article
Table 5. Table 5Fire test set results for different input configurations
配置 TP FP TN FN Acc Pre Recall F1-score
A 2 841 179 2 799 103 95.23% 94.07% 96.50% 95.26%
B 2 863 87 2 891 81 97.16% 97.05% 97.24% 97.13%
C 2 896 97 2 881 48 97.55% 96.76% 98.37% 97.56%

Table 6. Table 6Manual comparison and verification results between DNN algorithm and CTA
View table
View in Article
Table 6. Table 6Manual comparison and verification results between DNN algorithm and CTA
区域算法总计正确检测错误检测重合个数
中国中原 DNN 15 5 10 10
CTA 119 16 103
东南亚 DNN 86 76 10 51
CTA 58 53 5
美国西部 DNN 30 28 2 26
CTA 36 32 4

Tools

Get Citation

Copy Citation Text

Jinpeng CHEN, Lin SUN, Feifei XIE, Huijuan GAO, Shuai GE. Research on Fire Detection Method based on Deep Neural Network MODIS Data[J]. Remote Sensing Technology and Application, 2024, 39(4): 905

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites