Lightweight UAV detector achieves real time accuracy with reparameterized convolutions

Lightweight UAV detector achieves real time accuracy with reparameterized convolutions
by Riko Seibo
Tokyo, Japan (SPX) Sep 25, 2025

Remote sensing object detection is becoming increasingly important for UAV operations in disaster response, urban planning, and environmental monitoring. However, developing models that are both accurate and efficient on lightweight devices has remained difficult because UAV images capture objects of varying scales, orientations, and lighting under limited onboard computing power.

A research team from Osaka Metropolitan University, led by graduate student Hoang Viet Anh Le and Associate Professor Tran Thi Hong, has designed a new detection framework that tackles these constraints. Their approach centers on the Partial Reparameterization Convolution Block (PRepConvBlock), which lowers the complexity of convolutional operations while preserving strong feature extraction. This makes it possible to employ larger kernels that extend receptive fields and improve long-range feature interactions.

To further refine detection performance, the team developed a Shallow Bi-directional Feature Pyramid Network (SB-FPN). This design fuses information between shallow and deeper feature levels, enhancing visual representation and making object detection more robust in complex UAV imagery.

These components combine in a new architecture called SORA-DET (Shallow-level Optimized Reparameterization Architecture Detector). Tailored for UAV remote sensing, SORA-DET can employ up to four detection heads and balances speed with accuracy. Benchmark tests showed it achieved 39.3 percent mAP50 on the challenging VisDrone2019 dataset and 84.0 percent mAP50 on the SeaDroneSeeV2 validation set.

Despite its strong performance, the model remains compact and efficient, requiring 88.1 percent fewer parameters than standard one-stage detectors. It also demonstrated an inference speed as fast as 5.4 milliseconds, making it suitable for real-time applications on resource-limited UAV systems.

The researchers emphasize that SORA-DET opens possibilities for practical use in scenarios such as disaster management and search-and-rescue missions. By enabling accurate detection without heavy computational demands, the framework marks a step toward deploying AI-powered UAVs in critical field operations.

Research Report:Partial feature reparameterization and shallow-level interaction for remote sensing object detection