14 February 2024 in IEEE Transactions on Intelligent Vehicles
Road extraction is a powerful technique support to autonomous driving as it provides routable road information for motion planning algorithms. High-resolution aerial images offer comprehensive road information, facilitating the establishment of efficient and accurate road networks to monitor road changes in a timely manner. However, widespread occlusions and abundant local details pose challenges to highly accurate and continuous extraction, especially in areas with road bifurcations. To simultaneously improve both accuracy and connectivity of road extraction, in this paper, we propose a novel approach TPEGAN to integrate pixel-level segmentation and graph inference based on road pixel enhancement. By generating road pixels enhanced images, the generative adversarial network exploits the consistency among road pixels to embed pixel-level accuracy into the segmentation module. The multi-scale dual-branch segmentation module employs graph convolution reasoning to capture dependencies across different spatial regions, maintaining the connectivity of road networks. Extensive experiments on three public datasets demonstrate that TPEGAN outperforms SOTA methods with a considerable performance gap. As the complexity of road networks increases, the performance of TPEGAN degrades more slowly than SOTA method. Even in challenging urban scenes where the proportion of road pixels is more than 15%, TPEGAN retains its high performance and achieves a rIoU of 0.664 with an APLS of 72.81%, amounting to improvements of 4.1% and 2.62% over SOTA method, respectively.
25 July 2025 in IEEE Transactions on Geoscience and Remote Sensing
Accurate road extraction from remote sensing images is of great importance, yet the complex backgrounds, significant scale differences, and detailed characteristics of road information pose a tremendous challenge to the semantic modeling capabilities of deep learning methods. Recent studies have shown impressive success with diffusion-based methods in semantic segmentation. However, when applied to road extraction tasks, the remarkable performance can be diminished by intricate terrains and unreasonable feature fusion. In this article, we reframe road extraction as a mask-generating task, progressively refining road mask predictions through a timestep-by-timestep denoising process. Specifically, we propose a novel adaptive frequency-aware latent diffusion framework named AFLDiff. The pretrained VQ-VAE encoder with coupled perceptual compression suppresses semantic features that are weakly associated with roads in remote sensing images, reconstructing the image into a low-dimensional latent space for training. The adaptive frequency-aware module filters out redundant noise from the noisy features and enhances useful high- and low-frequency components, bridging the gap between noisy and semantic features for better fusion. Extensive experimental results demonstrate that our method achieves state-of-the-art (SOTA) performance on the DeepGlobe, Massachusetts, and CHN6-CUG datasets, with cross-dataset validation further confirming its robustness and generalization capabilities.
5 December 2025 in Engineering Structures