Detection-Friendly Nonuniformity Correction: A Union Framework for Infrared UAVTarget Detection

TL;DR

UniCD tackles infrared UAV target detection under nonuniform bias by jointly performing nonuniformity correction (NUC) and detection in an end-to-end framework. It models the bias field as a small set of polynomial coefficients predicted by a lightweight network, and enhances detection with a mask-supervised TEBS loss and a bias-robust BR loss that align corrected features with clear targets. A new IRBFD benchmark with 50,000 images (30k synthetic, 20k real) enables robust evaluation and generalization; experiments show UniCD achieves state-of-the-art performance while running in real time, demonstrating strong practicality for edge devices.

Abstract

Infrared unmanned aerial vehicle (UAV) images captured using thermal detectors are often affected by temperature dependent low-frequency nonuniformity, which significantly reduces the contrast of the images. Detecting UAV targets under nonuniform conditions is crucial in UAV surveillance applications. Existing methods typically treat infrared nonuniformity correction (NUC) as a preprocessing step for detection, which leads to suboptimal performance. Balancing the two tasks while enhancing detection beneficial information remains challenging. In this paper, we present a detection-friendly union framework, termed UniCD, that simultaneously addresses both infrared NUC and UAV target detection tasks in an end-to-end manner. We first model NUC as a small number of parameter estimation problem jointly driven by priors and data to generate detection-conducive images. Then, we incorporate a new auxiliary loss with target mask supervision into the backbone of the infrared UAV target detection network to strengthen target features while suppressing the background. To better balance correction and detection, we introduce a detection-guided self-supervised loss to reduce feature discrepancies between the two tasks, thereby enhancing detection robustness to varying nonuniformity levels. Additionally, we construct a new benchmark composed of 50,000 infrared images in various nonuniformity types, multi-scale UAV targets and rich backgrounds with target annotations, called IRBFD. Extensive experiments on IRBFD demonstrate that our UniCD is a robust union framework for NUC and UAV target detection while achieving real-time processing capabilities. Dataset can be available at https://github.com/IVPLaboratory/UniCD.

Figures (8)

• Figure 1: Three main categories of methods for UAV target detection in nonuniformity conditions. (a) Direct: detection models 2023YOLOv8Jocher are directly applied to nonuniformity degraded images. (b) Separate: correction model 2016IPTLiu serves as a pre-processing step, correcting images before passing them to detectors 2023YOLOv8Jocher. (c) Union: correction and detection are processed simultaneously in a unified framework. Previous methods solely concentrates on optimizing one task. Our UniCD concurrently emphasizes the joint enhancement of correction quality and detection accuracy.
• Figure 2: Overview of the proposed UniCD. Our UniCD integrates a bias field prediction network with an infrared UAV target detection network. These two components are fused into a unified pipeline and trained end-to-end. The target enhancement and background suppression (TEBS) loss is introduced to enhance UAV target features while suppressing the background. The bias robust loss is employed to balance correction and detection.
• Figure 3: Calculation of the proposed feature enhancement and background suppression (TEBS) loss.
• Figure 4: Construction of the bias-robust loss in the unified NUC and detection framework.
• Figure 5: P-R curves of our UniCD and other correction-then-detection methods on the IRBFD-syn. The area values under the curves are placed after the method names.