A Cross Entropy Based Deep Neural Network Model for Road Extraction from Satellite Images.

Bowei Shan,Yong Fang

doi:10.3390/e22050535

Abstract

This paper proposes a deep convolutional neural network model with encoder-decoder architecture to extract road network from satellite images. We employ ResNet-18 and Atrous Spatial Pyramid Pooling technique to trade off between the extraction precision and running time. A modified cross entropy loss function is proposed to train our deep model. A PointRend algorithm is used to recover a smooth, clear and sharp road boundary. The augmentated DeepGlobe dataset is used to train our deep model and the asynchronous training method is applied to accelerate the training process. Five salellite images covering Xiaomu village are taken as input to evaluate our model. The proposed E-Road model has fewer number of parameters and shorter training time. The experiments show E-Road outperforms other state-of-the-art deep models with 5.84% to 59.09% improvement, and can give the accurate predictions for the images with complex environment.

Highlights

In recent years, the great success of deep learning has influenced many areas
We compare our method with Deep Residual U-Net (DRUnet) [9] and D-LinkNet [13] to investigate the feasibility
We present a deep convolutional neural network model, E-Road, for road extraction from satellite images

Summary

Introduction

The great success of deep learning has influenced many areas. In remote sensing community, many problems, such as, understanding high spatial resolution satellite images, hyperspectral image analysis, SAR images interpretation, multimodal data fusion, and 3-D reconstruction have employed the deep learning technique. Thereafter, a series of improvement to DeepLab has been made [4,5,6,7,8], e.g., DeepLabv1/2/3/3+ The experiemnts of these models have demonstrated that DCNN-based algorithms are powerful tools for semantic image segmentation. This is because of the built-in invariance of DCNNs to local image transformations, which allows them increasingly to learn high level abstract data representations. Road extraction is a sub-problem of semantic image segmentation, in which only two types of objects, i.e., road or background, are considered. Under this framework, a lot of algorithms have been proposed by taking advangtage of deep learning models. Gao et al [15] devised a refined deep residual convolutional neural network (RDRCNN) framework with a postprocessing stage for road extraction

Methods

Results

Discussion

Conclusion