RASR-SASD3D: A Region-Adaptive Structure Recovery and Spatial-Adaptive Structure Distillation Method for LiDAR-Based 3D Object Detection in Rainy Weather

In rainy environments, rain-induced scattering, occlusion, and echo attenuation cause LiDAR point clouds to suffer from degraded observations, including spurious returns, missing valid points, and uneven spatial distributions. These degradations disrupt the local geometric structures of detected objects, resulting in a substantial decline in the performance of conventional 3D object detectors. Existing methods mainly rely on weather degradation simulation, input enhancement, or domain adaptation strategies, which are unable to recover object structures according to the degradation severity in different regions and do not fully exploit the stable cross-weather structural priors available in clear-weather point clouds. To address these issues, we propose a rainy-weather LiDAR 3D object detection method based on region-adaptive structure recovery and spatially adaptive structural distillation, termed RASR-SASD3D. The proposed method constructs a clear-weather teacher network and a rainy-weather student network, using the stable geometric and semantic representations learned from clear-weather point clouds as structural priors. First, a Region-Adaptive Structure Recovery (RASR) module is designed; it performs differentiated feature aggregation for near-range, middle-range, and far-range regions according to distance-dependent degradation severity and structural reliability. It leverages local neighborhood and contextual information to compensate for structural losses in rainy-day point clouds while suppressing noise responses. Second, we design a Spatially Adaptive Structural Distillation (SASD) module to transfer stable structural knowledge from the clear-weather teacher network to the rainy-weather student network at the levels of multiscale local features, spatial relations, and global scene representations. By imposing distance-adaptive constraints on valid spatial regions, the proposed module reduces the interference of rainy-weather noise, invalid responses, and spatial inconsistencies in point clouds during the distillation process. During training, the detection objective is jointly optimized with the structure recovery and structural distillation constraints, whereas only the rainy-weather student branch is retained during inference. Experimental results demonstrate that RASR-SASD3D effectively enhances the representation of object structures in rainy-weather point clouds, improves 3D object detection performance under different levels of rain-induced degradation, and exhibits strong robustness to adverse weather conditions.

Authors

Institutions

Publication Details

Journal
Sensors
Published
2026-11-16
DOI
https://doi.org/10.3390/s26227223
Primary Topic
Advanced Neural Network Applications
Type
article
Field-Weighted Citation Impact
0.00
Controls
|||
ALL TIME
JAN
FEB
MAR
APR
MAY
JUN
JUL
AUG
SEP
OCT
NOV
article

RASR-SASD3D: A Region-Adaptive Structure Recovery and Spatial-Adaptive Structure Distillation Method for LiDAR-Based 3D Object Detection in Rainy Weather

丁业辉, Yuntao Shi
Sensors
Advanced Neural Network Applications
article

RASR-SASD3D: A Region-Adaptive Structure Recovery and Spatial-Adaptive Structure Distillation Method for LiDAR-Based 3D Object Detection in Rainy Weather

丁业辉, Yuntao Shi
article en

Abstract

In rainy environments, rain-induced scattering, occlusion, and echo attenuation cause LiDAR point clouds to suffer from degraded observations, including spurious returns, missing valid points, and uneven spatial distributions. These degradations disrupt the local geometric structures of detected objects, resulting in a substantial decline in the performance of conventional 3D object detectors. Existing methods mainly rely on weather degradation simulation, input enhancement, or domain adaptation strategies, which are unable to recover object structures according to the degradation severity in different regions and do not fully exploit the stable cross-weather structural priors available in clear-weather point clouds. To address these issues, we propose a rainy-weather LiDAR 3D object detection method based on region-adaptive structure recovery and spatially adaptive structural distillation, termed RASR-SASD3D. The proposed method constructs a clear-weather teacher network and a rainy-weather student network, using the stable geometric and semantic representations learned from clear-weather point clouds as structural priors. First, a Region-Adaptive Structure Recovery (RASR) module is designed; it performs differentiated feature aggregation for near-range, middle-range, and far-range regions according to distance-dependent degradation severity and structural reliability. It leverages local neighborhood and contextual information to compensate for structural losses in rainy-day point clouds while suppressing noise responses. Second, we design a Spatially Adaptive Structural Distillation (SASD) module to transfer stable structural knowledge from the clear-weather teacher network to the rainy-weather student network at the levels of multiscale local features, spatial relations, and global scene representations. By imposing distance-adaptive constraints on valid spatial regions, the proposed module reduces the interference of rainy-weather noise, invalid responses, and spatial inconsistencies in point clouds during the distillation process. During training, the detection objective is jointly optimized with the structure recovery and structural distillation constraints, whereas only the rainy-weather student branch is retained during inference. Experimental results demonstrate that RASR-SASD3D effectively enhances the representation of object structures in rainy-weather point clouds, improves 3D object detection performance under different levels of rain-induced degradation, and exhibits strong robustness to adverse weather conditions.

SensorsVol. 26(22)
North China University of Technology (CN)
Advanced Neural Network Applications
AI Navigator

Ask Laika to Summarize, Analyze, and Connect papers live on the map.

Summarize Papers & Methodologies

Extract key findings, datasets, and comparative methods across publications.

Benchmark Rankings & Visual Analytics

Rank top research institutions, authors, funders, topics, and journals by Field-Weighted Citation Impact (FWCI) and paper volume with instant charts.

Connect Distant Disciplines

Bridge topological clusters on the map to find hidden collaborative intersections.

Laika access coming soon