Desertification Detection Using an Improved Variational Autoencoder-Based Approach Through ETM-Landsat Satellite Data

The accurate land cover change detection is critical to improve the landscape dynamics analysis and mitigate desertification problems efficiently. Desertification a challenging problem because of high degree similarity between some cases like-desertification phenomena, such as deforestation. This article provides an effective approach detect deserted regions based on Landsat imagery variational autoencoder (VAE). VAE model, deep learning-based has gained special attention in features extraction modeling due its distribution-free assumptions superior nonlinear approximation. Here, applied spectral signatures for detecting pixels affected by change. considered are extracted from multitemporal images include multispectral information, no prior image segmentation required. proposed method was evaluated publicly available remote sensing data using optical taken freely program. arid region around Biskra Algeria selected study area since it well-known that phenomena strongly influence this region. model compared with restricted Boltzmann machines, learning binary clustering algorithms, including Agglomerative, BIRCH, expected maximization, k-mean one-class support vector machine. comparative results showed consistently outperformed other models changes cover, mainly regions. also state-of-the-art algorithms.