Neural network for grain yield predicting based multispectral satellite imagery: comparative study

Abstract

Listen

Translate article

Estimations of crop yield predictions are vital in the management of agronomical matters. Such Agronomical issues affecting agriculture include agricultural management, national food policies, as well as the international crop trade-which is under the mandate of Food agriculture organization (FAO). Also, an increase in food demand due to the ever-growing population has contributed to the cultivation of large tracts of land. Thus has led to the evolution of diverse methods as well as systems deployed for prediction of crop yield including the application of satellite images. Satellite techniques are utilized due to their capacity to continuously cover large areas while providing accurate estimations of crop yields. In this context of crop yield estimations, the vegetation indices provided by the satellite sensors, as well as land surface variables such as weather elements, soil moisture, hydrological conditions, soil fertility, and fertilizer application is used. Where the convenience of data acquisition and high prediction accuracy is mandatory, many empirical models based on machine learning techniques were employed and the most successful methodology applied was the neural network. The neural network data input varied in the form of normalized histograms of a multi-spectral image bands, normalized vegetation index, absorbed active photosynthetic radiation, canopy surface, and environmental factors. Our findings indicate that the rapid advances in satellite technologies and ML techniques will provide affordable and comprehensive solutions for accurate grain prediction. Many remote sensing researches for yield estimation is needed to adjust and develop the existing methods for more accurate grain crop prediction.