Keywords: flood, multi-spectral, multi-sensor, texture feature, neural network
Abstract
In recent tens year, there have exist significant natural environment changes, especially in a country covered by monsoon climate as a tropical area. The environmental management, planning and monitoring must be done to avoid a recurrent disaster. Multi-temporal images are used to detect the environment changing, while the multi-spectral images provide necessary information for land cover interpretation. To conveniently assess the flooded area, radar images act as the flood monitoring due to the properties of all weathers, day-and-night and capability of cloud-piercing. Therefore, multi-sensors of such images are merged to hold each superior characteristics by using image fusion technique, as shown in this paper. The classification and interpretation for flooded area identification of fused images based on the texture analysis and neural network classification. These comprehensive method shows the efficient investigation and assessment the unupdated area.
1. Introduction
There are several efforts to monitor and assess flood area. Especially, the monsoon regions are suddenly inundated by flash flood caused by the storm and others natural hazard. Several techniques have been applied to estimate the flood area. To investigate and identify the damage areas.
The radar image uses the back-scattering wave technique in the antenna direction gives two evidence results, discontinuity/roughness of the object surface and the object absorption depending on the moisture. Therefore, SAR images are efficiently flood detection. However, the classification and interpretation of SAR image are quite difficult with low precision because its mechanism is quite different from multi-spectral image. Texture feature analysis in section 2 shows comprehensive methodology to describe such data.
As the multi-spectral images provide necessary information for land cover interpretation. The fusion details of an effective exploiting the complimentary nature of these different data types, especially the SAR- optical image have the clearly different characteristics. Therefore, the dominant data addition of multi-spectral image into SAR image will make the data analysis to be comfortable and easy.
Figure 1 JERS-1 SAR image acquired on June 03,1997, Surat Thani province.
Figure 2 JERS-1 SAR image acquired on August 30,1997.
This paper have been adopted this technique to perform the flood area classification. The mixed data of SAR and OPS (optical sensor) obtained from JERS-1 satellite in the flood period of central area of Thailand as study region. This region is water hole with flooding. JERS-1 SAR data acquired on June 03, 1997 and August 30, 1997 were taken before and during flood hazard from the tropical storm Zita, in Surat Thani province, were used. Because of its cloud penetration capability as shown in Fig.1 and 2, respectively. Fusion these images with JERS-1 OPS data acquired on March 14, 1997 were performed to distinguish flood area. Using the wavelet decomposition based on low wavelet coefficient of SAR image is included with OPS image, the wave absorption from the object moisture is combined with the other pysical data as mention in section 3.
Figure 3 JERS-1 OPS data acquired on March 14,1997.
To study the flood assessment, the image classification is used as a revolutionary computing methodology known as the multi-layer perception (MLP) neural network based on the back propagation (BP) algorithm suitably for multi-spectral image as shown in section 4. The fused image and texture content are applied to this mechanism. Thus, the classification and analysis of flood area will be high resolution and high accuracy. The results and conclusion are presented in last section.
