ANALYSIS OF LAND COVER CHANGES IN THE CITARUM WATESHED USING DEEP LEARNING ARCHITECTURES

• Main Author: Virdian, Ardhani
• Format: Final Project
• Language: Indonesia
• Online Access: https://digilib.itb.ac.id/gdl/view/73419
• Institution: Institut Teknologi Bandung

<p align="justify">Along with the increasing development of remote sensing technology, it opens up great opportunities in its utilization. One of its applications is for land cover classification. This classification is carried out to identify the types of land cover which can be used for various analyses, such as land cover change analysis. One method for land cover classification is using Deep Learning method. In the land cover classification process, satellite images and research area boundaries are required. The satellite image data used is Sentinel 2A imagery acquired from 2017 to 2020. This final project was created with the aim of performing land cover change analysis based on the results of the Deep Learning method and determining the accuracy level of several Deep Learning models used. The model used in the Deep Learning method is DeepLabV3 with ResNet50 and ResNet101 architectures. The land cover classification is divided into 4 classes, namely open land, built-up land, vegetation, and water. The accuracy of DeepLabV3+ResNet50 model is 92.64%, while DeepLabV3+ResNet101 model achieved 92.57% accuracy. The average overall accuracy for DeepLabV3+ResNet50 model is 83.56%, while for DeepLabV3+ResNet101 model is 81%. Based on these average overall accuracy results, the DeepLabV3+ResNet50 model outperforms the DeepLabV3+ResNet101 model for land cover classification in the study area used.