Land Cover Change Detection Using Deep Learning

Abstract

Land cover change detection is essential for monitoring environmental changes. This paper aims to address the need for detecting changes in land cover over large areas and datasets. The study conducts a design experiment using aerial photos of Yeka Subcity, Worda 10, in Addis Ababa, taken in 2018 and 2021. Our work represents a significant advancement in the efficient and precise analysis of large-scale data for land cover change detection. We propose an unsupervised learning approach that employs a Convolutional Autoencoder (CAE) to robustly learn features from the input data. Temporal variations are identified using Euclidean distance, and Otsu thresholding is applied to generate binary change maps. The CAE model, optimized with Mean Squared Error (MSE) as the loss function, achieved an 89% accuracy rate in detecting land cover changes. This deep learning-based approach demonstrates considerable promise and effectiveness for large-scale land cover change detection