Rice Grain Yield Estimation using Artificial Neural Network with Optical Satellite Imagery

Abstract

Estimation of rice yield is important for crop management food security trade and policy Visiting rice fields on a daily basis to check the quality of rice grain is time-consuming and labor sensitivity Instead many researchers utilized vegetation indices (VIs) from optical satellite images with a regression model to predict rice yield However the VIs are insufficient to delineate the features in the image and the relationship between indices and rice grain yield is generally nonlinear and complex which may result in an un-optimal estimation In this study texture features extracted by Grey-Level Co-occurrence Matrix (GLCM) are utilized in addition to VIs and Artificial Neural Network (ANN) is adopted to model the nonlinear relationship between extracted image features and the rice yields GLCM is able to extract texture features that represent shape edge and roughness of the paddy rice in the image The design of this ANN model contains input single or multiple hidden and output layers The input layer consists of 47 neurons with 4 original bands 11 VIs and 32 GLCM features Furthermore the best model of estimating the rice yield can be produced by adjusting and comparing the number of hidden layers and corresponding neurons in the ANN models Two SPOT-7 images acquired in two different days but similar stage of rice growing were used and a 10-fold cross validation was performed to evaluate the proposed model because of the limited in-situ samples The study area is Erlin Township Changhua County Taiwan that contains many rice fields Quantitative accuracy assessments were conducted to demonstrate the feasibility and performance of the proposed model in terms of Root Mean Square Error (RMSE) The experiments results show that the best ANN model of all models that have been tried is the ANN model that uses one hidden layer with 47 neurons It has a RMSE value of 0 6128 tons

Date of Award	2020
Original language	English
Supervisor	Chao-Hung Lin (Supervisor)