Deep Hybrid Network for Land Cover Semantic Segmentation in High-Spatial Resolution Satellite Images

Sultan Daud Khan,Louai Alarabi,Saleh Basalamah

doi:10.3390/info12060230

Sultan Daud Khan, Louai Alarabi + Show 1 more

Open Access

https://doi.org/10.3390/info12060230

Copy DOI

Abstract

Land cover semantic segmentation in high-spatial resolution satellite images plays a vital role in efficient management of land resources, smart agriculture, yield estimation and urban planning. With the recent advancement in remote sensing technologies, such as satellites, drones, UAVs, and airborne vehicles, a large number of high-resolution satellite images are readily available. However, these high-resolution satellite images are complex due to increased spatial resolution and data disruption caused by different factors involved in the acquisition process. Due to these challenges, an efficient land-cover semantic segmentation model is difficult to design and develop. In this paper, we develop a hybrid deep learning model that combines the benefits of two deep models, i.e., DenseNet and U-Net. This is carried out to obtain a pixel-wise classification of land cover. The contraction path of U-Net is replaced with DenseNet to extract features of multiple scales, while long-range connections of U-Net concatenate encoder and decoder paths are used to preserve low-level features. We evaluate the proposed hybrid network on a challenging, publicly available benchmark dataset. From the experimental results, we demonstrate that the proposed hybrid network exhibits a state-of-the-art performance and beats other existing models by a considerable margin.

Highlights

With the recent advancement in remote sensing technologies, such as satellites, drones, and airborne vehicles, etc., high-resolution satellite images are easy to acquire [1].This opens up new paradigms and research directions for the remote sensing community that offer different applications in diverse fields, for example, land cover segmentation [2,3,4], smart agriculture [5,6], traffic monitoring [7,8], disaster management [9], geo-localization [10], and urban planning [11,12]
Land cover classification and segmentation is an important application that extracts useful information about the type of land covered by agriculture, water, forest, urban, etc., which is crucial for land resource managers
To address the above mentioned limitations of existing deep learning networks, we proposed a hybrid network that consists of two deep neural network architectures, DenseNet [30] and U-Net [23]

Summary

Introduction

With the recent advancement in remote sensing technologies, such as satellites, drones, and airborne vehicles, etc., high-resolution satellite images are easy to acquire [1] This opens up new paradigms and research directions for the remote sensing community that offer different applications in diverse fields, for example, land cover segmentation [2,3,4], smart agriculture [5,6], traffic monitoring [7,8], disaster management [9], geo-localization [10], and urban planning [11,12]. With recent advances in computer vision and the success of deep neural networks with regard to optical natural images, several automated models [3,17] have been proposed in the literature that automatically perform semantic labeling (assign class) of land cover in high-resolution remote sensing images

Methods

Results

Conclusion