IJPAM: Volume 41, No. 4 (2007)


Tomohisa Konishi$^1$, Sigeru Omatu$^2$, Yuzo Suga$^3$
$^{1,2}$Department of Mathematics
Osaka Prefecture University
1-1 Gakuen-cho, Nakaku, Sakai, Osaka, 599-8531, JAPAN
$^1$e-mail: konishi@cadic.co.jp
$^2$e-mail: omatu@cs.osakafu-u.ac.jp
$^3$Earth Environment Information Center
Hiroshima Institute of Technology
2-1-1 Miyake, Saeki-ku, Hiroshima, 731-5193, JAPAN
e-mail: y.suga.mi@it-hiroshima.ac.jp

Abstract.The classification technique using neural networks has been recently developed. We apply a neural network of error back-propagation (BP) to classify remote sensing data including microwave and optical sensors for the estimation of rice-planted area. The method has capability of a nonlinear classification and the discrimination function can be determined by learning.

The satellite data were observed before and after planting rice. RADARSAT-1/SAR, ENVISAT-1/ASAR and SPOT-2/HRV data are used in Higashi-Hiroshima, Japan. Three images for RADARSAT and ENVISAT from April to June are used and one SPOT image is used for classification.

In case of the BP, the output layer has four clusters such as water region, urban area, rice-planted area and forest and the input data are the SAR data observed in three different seasons. Experimental results show that the present method is much better compared with classification of SAR image using the maximum likelihood (ML) method.

Received: September 12, 2007

AMS Subject Classification: 68U10

Key Words and Phrases: remote sensing, neural networks, classification of SAR

Source: International Journal of Pure and Applied Mathematics
ISSN: 1311-8080
Year: 2007
Volume: 41
Issue: 4