Deep neural network for semantic segmentation of satellite images

Abstract

Deep neural networks have become a crucial tool for satellite image processing, particularly in semantic segmentation tasks. This paper explores the use of deep neural networks for automated feature extraction and classification in Earth satellite images. It focuses on how deep architectures like U-Net and MobileNet handle multi-channel spectral data to achieve precise segmentation of various land covers and objects of interest. The paper discusses data preprocessing techniques, loss function selection, and optimization, along with examples of successful applications in mapping, agricultural monitoring, and urban planning. The study highlights the effectiveness of deep neural networks in addressing complex satellite image segmentation challenges and showcases their potential for future research and practical use in land management and environmental monitoring.