Identification of rice crops using Sentinel-l images

Abstract

Rice is among the main foods produced and consumed in Colombia. According to the 4th National Census of Rice Crops, there has been an increase in production that now reaches 2’971.975 tons at the national level. The objective of this study is to identify the rice fields in the municipality of Jamundí, Valle del Cauca, from 3 Sentinel 1 radar images between March and April 2017. The multilooking process was carried out. A medium filter of 7 x 7 was applied to reduce speckle. The geometric adjustment was performed with a 3-second SRTM digital elevation model (DEM). Subsequently, each image was calibrated radiometrically and the backscatter values (σ0) were obtained. Following the pre-processing, the three images were compiled and the study area was trimmed to minimize execution times. Four training vectors were defined: pasture, urban area, water and rice. From the vectors identified, a support vector machine (SVM) was trained with a radial basis function. Finally, the accuracy of the classification was validated by comparing the result with the land cover map of the Regional Autonomous Corporation of the Valley Cauca (CVC). Overall accuracy of 70%, user accuracy of 83% and producer accuracy of 64% were obtained. The kappa index was 0.55.