A Comparative Analysis of Gradient Boost Algorithm and Deep Neural Network for Land Cover Land Use Classification

Remote sensing satellite imagery has been considered a time-efficient way to analyze and understand some crucial components of typical urban ecosystems. Due to the fastdeveloping technology in the part of remote sensing images, so many applications are there using images collected from different satellite sources. Land cover analysis, land cover change identification, and mapping are important among them, for monitoring and prediction of natural calamities, disasters, etc. In the present study, comparing the performance of ML and DL techniques using Sentinel 1 and Sentinel 2 satellite images for the year 2020. Santa Catarina, Brazil was taken as the study area. Gradient boosting excels in handling gradient-based optimization to minimize loss functions, resulting in an ensemble model with improved accuracy and robustness. On the other hand, deep neural networks leverage their multi-layered architecture to learn complex feature representations, enabling them to capture intricate patterns in data and achieve superior performance in tasks such as image synthesis, language translation, and speech synthesis. Accuracy and Kappa analysis was done in both cases, and Score obtained from ML Model (Gradient Boost) in Google earth engine analysis is, Accuracy: 96.71, Kappa: 96.03, and from Deep Neural Network is Accuracy: 97.22, Kappa: 96.66. The detail study is useful for urban management and autonomous encroachment detection.