Wheat grain protein content estimation based on multi-temporal remote sensing data and generalized regression neural network

Abstract

Monitoring grain protein content in large areas by remote sensing is very important for guiding graded harvest, and facilitates grain purchasing for processing enterprises. Wheat grain protein content (GPC) at maturity was measured and multi-temporal Landsat TM and Landsat ETM + images at key stages in 2003, 2004 growth stages were acquired in this study. GPC was estimated with multi-temporal remote sensing data and generalized regression neural network (GRNN) method. Results show that the GPC prediction accuracy of the GRNN model is higher, with the average relative deviation of self-modeling, average relative deviation of cross-validation as 0.003%, 0.321%; 4.300%, 7.349% for 2003 and 2004 respectively. GRNN method proves to be reliable and robust to monitoring GPC in large areas by multi-temporal and multi-spectral remote sensing data. © 2012 IFIP International Federation for Information Processing.