Rice-Planted Area Extraction by RADARSAT Data by Competitive Neural Networks

The classification technique using the neural networks has been recently developed. We apply a competitive neural network of Learning Vector Quantization (LVQ) to classify remote sensing data including microwave and optical sensors for estimation of a rice field. The method has capability of a nonlinear discrimination function which is determined by learning. The satellite data were observed before and after planting rice in 1999. Three RADARSAT and one SPOT/HRV data are used in Higashi-Hiroshima City, Japan. RADARSAT image has only one band data, which is difficult to extract a rice field. However, SAR back-scattering intensity in a rice field decreases from April to May and increases from May to June. Thus, three RADARSAT images from April to June are used for this study. The LVQ classification was applied to RADARSAT and SPOT data in order to evaluate rice field estimation. The results show that the true production rate of rice field estimation for RADASAT data by using LVQ was approximately 60% compared with SPOT data. It is shown that the present method is much better compared with SAR image classification by the maximum likelihood (MLH) method.