Blur-Robust Detection via Feature Restoration: An End-to-End Framework for Prior-Guided Infrared UAV Target Detection

TL;DR

The paper tackles infrared UAV target detection under motion blur by proposing JFD3, an end-to-end dual-branch framework that learns feature-domain deblurring guided by a clear-image branch. It introduces a lightweight feature restoration module, a frequency structure guidance module to inject refined structure priors, and a feature-consistency self-supervised loss to align blurred representations with clean ones, achieving detection performance gains in both synthetic and real blur scenarios. Evaluation on the IRBlurUAV benchmark demonstrates superior accuracy and real-time efficiency, with thorough ablations confirming the contributions of FDD and FSGM. The approach offers practical impact for robust IR UAV surveillance under motion-induced degradation and introduces a new dataset to advance research in blur-robust infrared detection.

Abstract

Infrared unmanned aerial vehicle (UAV) target images often suffer from motion blur degradation caused by rapid sensor movement, significantly reducing contrast between target and background. Generally, detection performance heavily depends on the discriminative feature representation between target and background. Existing methods typically treat deblurring as a preprocessing step focused on visual quality, while neglecting the enhancement of task-relevant features crucial for detection. Improving feature representation for detection under blur conditions remains challenging. In this paper, we propose a novel Joint Feature-Domain Deblurring and Detection end-to-end framework, dubbed JFD3. We design a dual-branch architecture with shared weights, where the clear branch guides the blurred branch to enhance discriminative feature representation. Specifically, we first introduce a lightweight feature restoration network, where features from the clear branch serve as feature-level supervision to guide the blurred branch, thereby enhancing its distinctive capability for detection. We then propose a frequency structure guidance module that refines the structure prior from the restoration network and integrates it into shallow detection layers to enrich target structural information. Finally, a feature consistency self-supervised loss is imposed between the dual-branch detection backbones, driving the blurred branch to approximate the feature representations of the clear one. Wealso construct a benchmark, named IRBlurUAV, containing 30,000 simulated and 4,118 real infrared UAV target images with diverse motion blur. Extensive experiments on IRBlurUAV demonstrate that JFD3 achieves superior detection performance while maintaining real-time efficiency.

Figures (4)

• Figure 1: Three strategies for UAV target detection under motion blur. (a) Direct: The detector directly processes blurred images. (b) Separate: Image-domain deblurring serves as a preprocessing step before detection. (c) Joint: Feature-domain deblurring and detection are simultaneously addressed in an end-to-end framework. Our JFD3 jointly handles both tasks and leverages structural priors from the deblurring network to enhance the feature representation of the detection network.
• Figure 2: Overview of the proposed JFD3, which first enhances degraded features through feature-domain restoration and then refines structural information using the frequency structure guidance module. The clear image branch supervises the blurred image branch using feature restoration loss and feature consistency self-supervised loss.
• Figure 3: Overview of FSGM. The FFRB processes prior through high-pass filtering and attention mechanisms to refine feature representations. The SPIB integrates refined structure prior into the feature map.
• Figure 4: Comparison of detection results on IRBlurUAV-syn and IRBlurUAV-real, including direct, separate, and joint detection methods. Green and red boxes represent ground-truth and detected targets, respectively. Close-up views are shown in the bottom-right corner.