Chunle Miao

Road segmentation is one of the essential tasks in remote sensing. Large-scale high-resolution remote sensing images originally have larger pixel sizes than natural images, while the existing models based on Transformer have the high computational cost of square complexity, leading to more extended model training and inference time. Inspired by the long text Transformer model, this paper proposes a novel hybrid attention mechanism to improve the inference speed of the model. By calculating several diagonals and random blocks of the attention matrix, hybrid attention achieves linear time complexity in the token sequence. Using the superposition of adjacent and random attention, hybrid attention introduces the inductive bias similar to convolutional neural networks (CNNs) and retains the ability to acquire long-distance dependence. In addition, the dense road segmentation result of remote sensing image still has the problem of insufficient continuity. However, multiscale feature representation is an effective means in the network based on CNNs. Inspired by this, we propose a multi-scale patch embedding module, which divides images by patches with different scales to obtain coarse-to-fine feature representations. Experiments on the Massachusetts dataset show that the proposed HA-RoadFormer could effectively preserve the integrity of the road segmentation results, achieving a higher Intersection over Union (IoU) 67.36% of road segmentation compared to other state-of-the-art (SOTA) methods. At the same time, the inference speed has also been greatly improved compared with other Transformer based models.

Recent advances in deep learning have shown remarkable performance in road segmentation from remotely sensed images. However, these methods based on convolutional neural networks (CNNs) cannot obtain long-range dependency and global contextual information because of the intrinsic inductive biases. Motivated by the success of Transformer in computer vision (CV), excellent models based on Transformer are emerging endlessly. However, patches with a fixed scale limit the further improvement of the model performance. To address this problem, a dual-resolution road segmentation network (DCS-TransUperNet) with a features fusion module (FFM) was proposed for road segmentation. Firstly, the encoder of DCS-TransUperNet was designed based on CSwin Transformer, which uses dual subnetwork encoders of different scales to obtain the coarse and fine-grained feature representations. Secondly, a new FFM was constructed to build enhanced feature representation with global dependencies, using different scale features from the subnetwork encoders. Thirdly, a mixed loss function was designed to avoid the local optimum caused by the imbalance between road and background pixels. Experiments using the Massachusetts dataset and DeepGlobe dataset showed that the proposed DCS-TransUperNet could effectively solve the discontinuity problem and preserve the integrity of the road segmentation results, achieving a higher IoU (65.36% on Massachusetts dataset and 56.74% on DeepGlobe) of road segmentation compared to other state-of-the-art methods. The considerable performance also proves the powerful generation ability of our method.