Skip to content
Open access

A Lightweight Scale-Adaptive Detector for Small UAV Detection in Complex Scenarios

Nov 2026 · IEEE Sensors Journal · Vol 26, pp. 33144-33160 · 0 citations · 61 references

Abstract

Distant uncrewed aerial vehicles (UAVs) usu-ally appear with small object sizes, weak textures, blurry boundaries, and severe background interference, which poses significant challenges for object detection in anti-UAV scenarios. Existing efficient detectors still suffer from feature degradation during repeated downsampling, and fixed receptive-field feature fusion is insufficient to handle the large scale variations caused by different target distances. To address these problems, this article proposes a lightweight scale-adaptive detector for anti-UAV small-object detection. Specifically, a Micro-target Contrast Enhancement (MCE) module is designed to reorganize shallow features by jointly modeling original appearance information, local contrast residuals, and surrounding contextual cues, thereby strengthening weak UAV responses in shallow high-resolution features. A Scale-adaptive Multireceptive-field Feature Fusion (SMFF) module is further proposed to perform location-dependent fusion among complementary local and contextual sampling patterns through branch-wise normalized weights, improving the adaptability of the detector to UAVs at different scales. Moreover, a detail-preserving adaptive downsampling (DPAD) module is developed to reduce small-object information loss during downsampling by adaptively integrating a detail branch and an anti-aliasing branch. By jointly organizing shallow local-contrast and contextual cues, adaptively allocating complementary local and contextual observations, and preserving fine-grained details during feature propagation, the proposed detector better handles distant UAV detection in complex backgrounds. Experiments on the Dalian University of Technology (DUT)-Anti-UAV and Anti-UAV (Inf) datasets demonstrate that the proposed method improves detection accuracy and small-object performance while maintaining a lightweight model structure and competitive computational cost.

Read PDF

We use cookies to run the site and, with your consent, for analytics and to show ads. See our Cookie Policy.