|
Image Processing
|
Flooded Area Assessment with Fused Multi-Spectral
Multi-Sensor by using Texture Feature Analysis and
Neural Network Classification
5. Methodology And Results.
5.1 Method
Pre-processing/image preparation
- The SAR data obtains 16 bit and are reduced to 8 bits in order to obtain 256 values of intensity by using linear scaling.
- From wavelet decomposition [5], the low wavelet coefficient of SAR images will be used within two reasons, to remove the speckle noise and continue the proper data for applying to neural network training algorithm.
- The 12.5m x 12.5m resolution of SAR data were reduced to 25m x 25m in the same order of OPS resolution. All image should be registered and geometric corrected.
Texture content extraction
- Follow the instructions as in section 2 with P-operator "one pixel to the right and one pixel below", co-occurrence matrix can be constructed.
- Using statistical parameters, only second-order element difference moment and entropy as a set of descriptors to reduce input node of MLP neural network and save calculation time. The first descriptor represents the value of the variogram at a lag distance as a P-position operator that is applied to develop the co-occurrence matrix. The second descriptor is a measure of randomness, achieving its highest value while all elements of co-occurrence matrix are equal.
Images fusion and classification
- Data fusion technique as mentioned in section 3, is used as shown in Fig.5.
- After pre-processing satellite image prepares then applying to neural network
- classification. Two kind of texture content will be added in to two last input nodes of network to increase classification accuracy.
5.2 Results
Table 1. The result of flood assessment by neural network classification with data fusion.
Classifi-
cation |
Water |
Cloud |
Urban |
Vegetation |
Bare Soil |
| Flood |
Non-Flood |
Flood |
Non-Flood |
Flood |
Non-Flood |
Result/
testing
(pixel) |
478/500 |
47/50 |
45/50 |
43/50 |
44/50 |
90/100 |
44/50 |
89/100 |
Corre-
ction |
95.6 |
94.0 |
90.0 |
86.0 |
88.0 |
90.0 |
88.0 |
89.0 |
Figure 5 Fused image for flood area assessment.
6. Conclusion
As results, above method show effectiveness and efficiency of multi-temporal SAR data, whichare very useful for flood assessment and monitoring. The SAR texture content helps us to identify a flood area. While as the OPS data provide the necessary information for land cover interpretation with the highly reliable result.
Acknowledgement
The ground truth informations are kindly supported by Office of Agricultural Economics, Ministry of Agriculture, Royal Thai Government.
References
- Xiaomei, Y., and Chenghu, Z., 1998. Recognition of Flooded Area in Radar Image Using
- Texture Feature Analysis. Proc. of 19 th Asian Conference on Remote Sensing (ACRS'98), Manila, Philipines, pp. P22-1 - P22-6.
- Carr, J. R., 1998. The Semivariogram in Comparison to the Co-Occurrence Matrix for
- Classification of Image Texture. IEEE Trans. Geoscience and Remote Sensing, Vol.36, No.6, pp.1945-1952.
- Wisetphanichkij, S., Dejhan, K., Cheevasuvit, F., Mitatha, S., Arungsrisangchai, I., Yimman, S., Pienvijarnpong, C., Soonyeekan, C. and Chanwutitum, J., 1999. A fusion approach of multispectral with SAR image for flood area analysis. Proc.of the 20 th Asian Conference on Remote Sensing (ACRS'99), Hong Kong, China, pp.53-58.
- Wisetphanichkij, S., Dejhan, K., Cheevasuvit, F., Mitatha, S., Netbut, C., Pienvijarnpong, C., Soonyeekan, C. and Chanwutitum, J., 1999. Multi-temporal cloud removing based on image fusion with additive wavelet decomposition Proc. of the 20th Asian Conference on Remote Sensing (ACRS'99), Hong Kong, China, pp.1109-1114.
|
|
|
|
|
