Estimation of Rice Area from Satellite Data Using Neural Networks

It is well-known that a neural network is useful to classify several patterns. In order to estimate the rice area we apply a network of learning vector quantization (LVQ) to remote sensing data including Synthetic Aperture Radar (SAR) and optical sensors for estimation of a rice area. The satellite data were observed before and after planting rice. Three RADARSAT and one SPOT/HRV data are used in Higashi-Hiroshima City, Japan. RADARS AT image has only one band data and it is difficult to extract a rice area. However, SAR back-scattering intensity in a rice area decreases from April to May and increases from May to June. Thus, three RADARS AT images from April to June are used to know the changes of rice growth. The LVQ classification is applied to RADARS AT and SPOT data in order to evaluate rice area. It is shown that the true production rate of rice area can be estimated from RADASAT data using LVQ by approximately 60% compared with SPOT data. It will be shown that the proposed method is much better compared with SAR image classification by the maximum likelihood (MLH) method.