Unmanned Aerial Vehicle(UAV) remote sensing has been widely adopted across various domains, where accurate detection of high-density small objects remains a critical challenge for low-altitude intelligent perception. We propose FGUDet, a novel framework that explores fine-grained distribution optimization for bounding box regression of small objects. FGUDet comprises two key components: Scale-Aware Fine-grained Distribution Refinement (SA-FDR) and Area-Weighted Localization Self-Distillation(AW-LSD). SA-FDR employs a dynamic probability residual iteration mechanism to achieve sub-pixel-level bounding box modeling, significantly enhancing robustness against blur in small objects. AW-LSD dynamically allocates knowledge transfer weights based on object size. This compels the shallow decoders to prioritize learning the localization patterns of small objects while suppressing interference from background noise. Furthermore, we introduce a Spectrum-Modulated Feed-forward Network (SMFFN) to decouple and enhance the low-frequency semantics and high-frequency details within feature maps, thereby mitigating the loss of edge texture in low-resolution objects. The lightweight FGUDet-N achieves 21.9% AP95 on the VisDrone dataset with only 3.7M parameters. FGUDet-X attains a breakthrough performance of 40.9% AP95, substantially outperforming state-of-the-art UAV-based detectors. Extensive evaluations on the UAVDT dataset further demon-strate the superior detection capability of FGUDet in UAV scenarios.

Millimeter-Wave radar has been widely applied in the field of autonomous driving due to an excellent performance under complex weather conditions. However, in practical roadside scenarios, the challenge of sparse point clouds leading to clustering difficulties and the issue of large vehicle point clouds dispersing, resulting in fragmentation, currently hampers the practical ap-plication of radar sensors. We propose an adaptive point cloud clustering algorithm based on DBSCAN. First, we propose an improved DBSCAN clustering algorithm based on distance and speed thresholds, which enhances the differentiation of point clouds between different vehicles, and an adaptive ellipse gate strategy to solve the large vehicle point clouds fragmentation problem. Then, a secondary clustering algorithm based on azimuth is exploited, effectively addressing the issues of large vehicle fragmentation and anomalous speed values. Practical roadside experimental results demonstrate that our proposed algorithm significantly outperforms traditional algorithms, showing considerable potential in practical applications.