USE OF UAV IMAGES FOR PEATLAND COVER CLASSIFICATION USING THE CONVOLUTIONAL NEURAL NETWORK METHOD

Abstract

Land cover or vegetation density in tropical peatland is an essential factor in hydrology response in geographic analysis, ranging from physical geography studies and approaches to sustainable planning to environmental research. Vegetation analysis according to the Indonesian National Standard (SNI 7645:2014) is classified on the basis of density. The vegetation density index is divided into four categories: non-vegetation, bare, medium, and high. In the technical aspect, to obtain information related to vegetation, this can be done using remote sensing. Remote sensing uses two types of data to obtain information: satellite data and UAV data. This study used UAV data with shooting locations in the Liang Anggang Protection Forest for classifying land cover. The method used was convolutional neural network with feature extraction used in this study was GLCM. This research used the ShuffleNet v2 architecture for the CNN method. The findings of this study used two models: the CNN model without GLCM process and compared to the CNN model with the addition of GLCM process, resulting in a comparison that was quite far from the accuracy value obtained. The CNN model obtained an accuracy value of 80%, while the CNN model with GLCM using segmentation gained 49.9% and without segmentation - 44.53%. Keywords: Tropical Peatland, Vegetation Density, Classification, Class, Convolutional Neural Network, Gray Level Co-Occurrence Matrix, Accuracy DOI： https://doi.org/10.35741/issn.0258-2724.58.3.63