In recent years, the advances in high-resolution satellite imagery have led to the popularity of automatic road extraction. However, most existing methods suffer from high computational cost and low efficiency. In this paper, we propose two novel encoder-decoder deep networks to tackle the automatic road extraction problem. The proposed methods integrate Atrous Spatial Pyramid Pooling (ASPP) and Dense Convolutional Network (DenseNet) on Unet. We implement our proposed models on DeepGlobe dataset and Massachusetts road extraction dataset. The experimental results show that our model is computationally efficient and able to effectively extract multi-scale global features and to preserve spatial information from deeper networks.
Paper
Full text
Deep Convolutional Neural Networks for Road Extraction
Semantic Scholar · Computer Science · 2020
Abstract
In recent years, the advances in high-resolution satellite imagery have led to the popularity of automatic road extraction. However, most existing methods suffer from high computational cost and low efficiency. In this paper, we propose two novel encoder-decoder deep networks to tackle the automatic road extraction problem. The proposed methods integrate Atrous Spatial Pyramid Pooling (ASPP) and Dense Convolutional Network (DenseNet) on Unet. We implement our proposed models on DeepGlobe dataset and Massachusetts road extraction dataset. The experimental results show that our model is computationally efficient and able to effectively extract multi-scale global features and to preserve spatial information from deeper networks.