Selection of the Most Suitable Sentinel-2 Bands and Vegetation Index for Crop Classification By Using Artificial Neural Network (Ann) and Google Earth Engine (Gee)

Abstract

The aim of this study is to determine the most suitable date, band and vegetation indices for the crop classification by using the Artificial Neural Networks method in irrigated agricultural areas and the usage possibilities of the web-based remote sensing platform GEE. For this purpose, the crop pattern of Cifteler irrigation scheme is located in Eskisehir, Turkey has been investigated. During the study, GEE has shown ability as an extremely powerful tool for obtaining high-resolution Sentinel 2 satellite images on a very wide area, for processing these images and for the use and export of results in classification study. It was seen that, because of GEE is cloud-based, it performs tasks quickly and conveniently with a large number of servers without depending on the performance of local computers. 12 band and 21 vegetation index values for each of the irrigation parcels were transferred to the MS Excel for use in the JMP statistical program. During the study, classification was performed with Artificial Neural Networks. During the classification process, Generalized R-2 and overall accuracy calculated respectively 96% and 94%.