Abstract:Airborne array-interferometric synthetic aperture radar (array-InSAR), one of the implementation methods of tomographic SAR (TomoSAR), has the advantages of all-time, all-weather, high consistency, and exceptional timeliness. As urbanization continues to develop, the utilization of array-InSAR data for building detection holds significant application value. Existing methods, however, face challenges in terms of automation and detection accuracy, which can impact the subsequent accuracy and quality of building modeling. On the other hand, deep learning methods are still in their infancy in SAR point cloud processing. Existing deep learning methods do not adapt well to this problem. Therefore, we propose a Position-Feature Attention Network (PFA-Net), which seamlessly integrates positional encoding with point transformer for SAR point clouds building target segmentation tasks. Experimental results show that the proposed network is better suited to handle the inherent characteristics of SAR point clouds, including high noise levels and multiple scattering artifacts. And it achieves more accurate segmentation results while maintaining computational efficiency and avoiding errors associated with manual labeling. The experiments also investigate the role of multidimensional features in SAR point cloud data. This work also provides valuable insights and references for future research between SAR point clouds and deep learning.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is how to improve the degree of automation and detection accuracy in the semantic segmentation of urban building point clouds. Specifically, existing methods face challenges in terms of automation and detection accuracy, which may affect the accuracy and quality of subsequent building modeling. Moreover, although deep - learning methods have made certain progress in other fields, they are still in the initial stage in SAR point - cloud processing, and the existing deep - learning methods are not fully adapted to this problem. Therefore, the paper proposes a method based on the Position - Feature Attention Network (PFA - Net), aiming to seamlessly integrate position encoding and point transformers to better handle the building target segmentation task in SAR point clouds. ### Main Problems 1. **Automation and Detection Accuracy**: Existing methods have deficiencies in automation and detection accuracy, which affects the quality of subsequent building modeling. 2. **Adaptability of Deep - Learning Methods**: Existing deep - learning methods perform poorly when dealing with SAR point clouds, and new methods are required to improve adaptability. 3. **High Noise and Multipath Scattering**: SAR point - cloud data has a high noise level and multipath scattering artifacts, and effective methods are needed to handle these characteristics. ### Solutions The paper proposes PFA - Net, a deep - learning model that combines position encoding and point transformers, aiming to directly process SAR point - cloud data and achieve point - cloud segmentation of building facades and roofs. The main features of PFA - Net include: - **Position - Feature Attention Mechanism**: Effectively extract and aggregate multi - dimensional features through the position encoding block, feature transformation module, and attention pooling layer. - **Multi - scale Sampling and Local Neighborhood Grouping**: Gradually expand the feature receptive field through multi - scale sampling and local neighborhood grouping methods, and finally extract global features. - **Upsampling and Feature Fusion**: Based on the extracted global and local features, combined with the downsampled point features, perform upsampling to obtain the segmentation results of all point clouds. ### Experimental Results The experimental results show that PFA - Net can better handle the inherent characteristics of SAR point clouds, including high noise levels and multipath scattering artifacts, and achieve more accurate segmentation results while maintaining computational efficiency. In addition, the experiment also explores the role of multi - dimensional features in SAR point - cloud data, providing a valuable reference for future research. ### Conclusions By proposing PFA - Net, the paper solves the problems of automation and detection accuracy in SAR point - cloud building target segmentation and provides new ideas for the application of deep - learning methods in SAR point - cloud processing.

Position-Feature Attention Network-Based Approach for Semantic Segmentation of Urban Building Point Clouds from Airborne Array Interferometric SAR

Building point cloud reconstruction in TomoSAR based on deep learning semantic segmentation

Semantic Segmentation of Urban Airborne LiDAR Point Clouds Based on Fusion Attention Mechanism and Multi-Scale Features

Semantic segmentation of 3D indoor LiDAR point clouds through feature pyramid architecture search

SAR Image Segmentation Based on Hierarchical Visual Semantic and Adaptive Neighborhood Multinomial Latent Model

PSFE-Net: Semantic Segmentation Network for Airborne LiDAR Transmission Corridor Scenes Inspection

SARNet: Semantic Augmented Registration of Large-Scale Urban Point Clouds

Building detection in SAR images based on fusion of classic and deep learning features

MAFF-HRNet: Multi-Attention Feature Fusion HRNet for Building Segmentation in Remote Sensing Images

Semantic Segmentation and Roof Reconstruction of Urban Buildings Based on LiDAR Point Clouds

LFEA-Net: semantic segmentation for urban point cloud scene via local feature extraction and aggregation

Multilevel intuitive attention neural network for airborne LiDAR point cloud semantic segmentation

Multispectral LiDAR Point Cloud Segmentation for Land Cover Leveraging Semantic Fusion in Deep Learning Network

A2-FPN for Semantic Segmentation of Fine-Resolution Remotely Sensed Images

Buildings Detection in VHR SAR Images Using Fully Convolution Neural Networks

NPSFF-Net: Enhanced Building Segmentation in Remote Sensing Images via Novel Pseudo-Siamese Feature Fusion

DFAMNet: dual fusion attention multi-modal network for semantic segmentation on LiDAR point clouds

Building extraction from oblique photogrammetry point clouds based on PointNet++ with attention mechanism

Semantic segmentation of urban land classes using a multi-scale dataset

Attention feature fusion awareness network for vehicle target detection in SAR images

Hybrid CNN-LSTM Architecture for LiDAR Point Clouds Semantic Segmentation