Patch based Classification using ResNet for Land Cover changes detection of Batu City

Abstract

The purpose of this study is to analyze the variations in land cover in Batu City, East Java Province, Indonesia, utilizing a patch-based classification strategy and deep learning. This study provides a preliminary estimation of land cover change in Batu City. The research also highlights the possibility of using deep learning techniques to analyze land use and land cover (LULC) variations in other urban areas with greater precision and efficiency. The EuroSAT dataset is used to train a classification model for patch labeling using the ResNet-50 architecture. Comparing the land cover of Batu City in 2001 and 2022 allows us to detect LULC changes, with almost 50% of the patch changing. The results indicate that ‘Housing’ and ‘Road’ become the most changed categories, while the vegetation areas decrease in number. The results demonstrate that the ResNet-50 architecture is capable of classifying patches and detecting LULC changes with an accuracy of 88% and an execution time of approximately 126.53 seconds.