Sentinel-1A image classification for identification of garlic plants using decision tree and convolutional neural network

Abstract

The Indonesian government launched a garlic self-sufficiency program by 2033 to reduce imports by monitoring garlic lands in several central garlic areas. Remote sensing using satellite imageries can assist the monitoring program by mapping the garlic lands. A previous study has classified Sentinel-1A satellite imageries to identify garlic lands in Sembalun Lombok Indonesia using the decision tree C5.0 algorithm with three scenarios data input and produced a model with an accuracy of 78.45% using scenarios with two attributes vertical-vertical (VV) and vertical-horizontal (VH) bands. Therefore, this study aims to improve the accuracy of the classification model from the previous study. This study applied two classification algorithms, decision tree C5.0 and convolutional neural network (CNN), with two new scenarios which used two new combinations of attributes). The results show that the use of new data scenarios as input for C5.0 can not increase the previous model's accuracy. While the use of the CNN algorithm shows that it can improve the previous study's accuracy by 7.91% because it produced a model with an accuracy of 86.36%. This study is expected to help garlic land identification in the Sembalun area to support government programs in monitoring garlic lands.