![Hier klicken, um den Treffer aus der Auswahl zu entfernen](images/unchecked.gif) |
Titel |
A Spatio-Temporal Model for Forest Fire Detection Using MODIS Data |
VerfasserIn |
Jing Li, Adu Gong, Yanling Chen, Jingmei Wang |
Konferenz |
EGU General Assembly 2017
|
Medientyp |
Artikel
|
Sprache |
en
|
Digitales Dokument |
PDF |
Erschienen |
In: GRA - Volume 19 (2017) |
Datensatznummer |
250150224
|
Publikation (Nr.) |
EGU/EGU2017-14661.pdf |
|
|
|
Zusammenfassung |
Contextual algorithm and Muti-temporal analysis are currently the most widely used in fire
detection based on remote sensing technology. However, muti-temporal analysis ignores the
correlation between the inspected pixel and its neighboring pixels (spatial heterogeneity)
(Equation (1)). Contextual algorithm only focuses on a single scene, and ignores the internal
differences of the background pixels, which increases the commission error. Due to the
muti-temporal analysis and contextual algorithm are used for different processes of
fire detection, the combination between them will increase the accuracy for fire
detection.
BTti1- BT-it2- BT-it3
BTto1 = αBT ot2= βBT ot3=...
(1)
Where BTtin is the bright temperature (BT) of the valid neighboring pixel i of the
inspected pixel o at time tn ,(i=1,2,…,N), N is the number of the valid neighboring
pixels(Which depends on the condition of context), BT otn is the BT of o at time
tn .
In this paper, We coupled the muti-temporal analysis with contextual algorithm
and proposed a region-adaptive spatio-temporal model for forest fire detection: (1)
Pre-processing: Cloud, water, potential background fires and bright fire-free targets masking
(refer to the context method); (2) Adjust the threshold for identifying potential fire-points for
different study areas (Equation 2); (3) The spatial relationship of BT between the
inspected pixels and its neighboring pixels in current time is build based on the spatial
relationship of BT between them in the multiple previous images, and the BT of the
inspected pixels is estimated based on the present spatial relationship and the BT of its
neighboring pixels and using inverse distance weighted method (Equation 3). (4) The
predicted BT value of the inspected pixel at a certain time is the weighted sum
of the value obtained by (3) and the real BT value of the inspected pixel at the
previous time (Equation 4); (5) Relative fire pixels judgment(refer to the context
method).
BT4 > BT4S,DBT > 10k,ρ0.86< 0.3
(2)
Where BTj ,ρj are the BT and the reflectance in channel j respectively,
DBT = BT4 − BT11 ,BT4S is the background BT in whole study area.
o ∑N
^BT tn = [Ftin × W itn × BT itn]
i=1
(3)
Where i BTotn- i i i
Ftn = a× BTitn + (1− a)× Ftn−1,a ∈ [0,1],Ft1 = 1,W tn is the inverse
distance weighted coefficient of the valid neighboring pixel i at time tn , o
^BT tn is the
predicted BT value of the inspected pixel o at time tn , this equation applies to 4-band and
11-band.
--- o
BT otn = b ×B^T tn + (1− b)× BT otn−1,b ∈ [0,1]
(4)
Where BTton−1 is the real BT of the inspected pixel o at time tn , ---o o
BT t1 = ^BT t1 .
The proposed model was applied to forest fires in Victoria(Australia) in 2009, southern
California(America) in 2016 and Fort McMurray(Alberta, Canada) in 2016, using MODIS
time-series images. The proposed algorithm described in this paper can significantly reduce
the commission error and detect more fire pixels than MOD14 using Landsat-8 images as a
reference dataset. The spatio-temporal model make full use of spatial information and
temporal information and is proved to be more effective than the optimized contextual
algorithm.
Keywords: Fire detection; Spatio-temporal model; MODIS |
|
|
|
|
|