What is it about?
Road extraction from remote sensing images has attracted widespread attention of researchers due to its crucial role in the fields of autopilot, urban planning, navigation and other fields. However, the task becomes challenging as the roads in remote sensing images are easily occluded by obstacles such as shadows, buildings and trees. In this Letter, a Cascade Fusion Network for Road extraction (CFRNet) in remote sensing images is proposed. Considering the lightweight characteristics of MobileNet Block (MbBlock), it’s utilized as the feature extraction module of the backbone network. To enable CFRNet to generate and fuse more features at multi-scale, we design several cascade stages. Each stage includes a Sub-backbone for feature extraction and a Triple-level Adaptive Feature Fusion module (TAFF) for feature fusion. This structure can more deeply and effectively fuse multi-scale features with most of the parameters in the entire backbone. The experimental results demonstrate that the proposed CFRNet significantly outperforms other state-of-the-art methods on the publicly available Istanbul City Road Dataset and DeepGlobe Road Dataset. Specifically, it achieves an Intersection over Union (IoU) of 89.76%, reflecting a 5.3% improvement on the Istanbul dataset, and 67.22% with a 0.98% enhancement on the DeepGlobe Road Dataset.
Featured Image
Photo by Denys Nevozhai on Unsplash
Why is it important?
In this letter, A cascade fusion network is proposed to extract roads from remote sensing images. • Several cascade stages are designed to generate and fuse more multi-scale feature information. Each Stage includes a sub-backbone network for multi-scale feature extraction and a TAFF for feature fusion. • The proposed CFRNet outperforms the state-of-the-art networks on two large-scale public datasets. And the code has been made public.
Perspectives
In this letter, we propose a Cascade Fusion Network for Road extraction in remote sensing images. Through cascade structure and adaptive feature fusion module, we can more effectively utilize multi-scale features to extract roads from remote sensing images. And we have achieved results surpassing other state-of-the-art methods on the two public available datasets. Although CFRNet can effectively predict roads by extracting and fusing multi-scale features, there are still certain shortcomings, such as the feature extraction module MbBlock not being a multi-scale feature extraction module. If the MbBlock module can be replaced with a multi-scale feature extraction module, it may have better results. Moving forward, we intend to incorporate the Transformer [22] and multiscale features into our feature extraction module, which will bolster the backbone’s capacity to capture global features. Simultaneously, we will simplify the network to make it suitability for deployment in projects related to road extraction.
Haoqi Wang
Beijing Information Science and Technology University
Read the Original
This page is a summary of: CFRNet: Road Extraction in Remote Sensing Images Based on Cascade Fusion Network, IEEE Geoscience and Remote Sensing Letters, January 2024, Institute of Electrical & Electronics Engineers (IEEE),
DOI: 10.1109/lgrs.2024.3409758.
You can read the full text:
Contributors
The following have contributed to this page
