In the realm of infrared imaging and detection, a team of researchers from the University of Electronic Science and Technology of China, led by Hongyang Xie and Hongyang He, along with Victor Sanchez, has developed a novel approach to improve the detection of small targets in infrared images. Their work, titled “TAPM-Net: Trajectory-Aware Perturbation Modeling for Infrared Small Target Detection,” addresses a persistent challenge in the field of infrared small target detection (ISTD), which is crucial for various applications including surveillance, search and rescue, and military operations.
Infrared small target detection is fraught with difficulties due to weak signal contrast, limited spatial extent, and cluttered backgrounds. While convolutional neural networks (CNNs) and Vision Transformers (ViTs) have shown promise in enhancing detection performance, current models struggle to trace how small targets cause directional changes in the feature space, which is vital for distinguishing actual signals from background noise.
The researchers propose a new model called the Trajectory-Aware Mamba Propagation Network (TAPM-Net). This model explicitly tracks the spatial diffusion behavior of feature disturbances induced by targets. TAPM-Net consists of two key components: the Perturbation-guided Path Module (PGM) and the Trajectory-Aware State Block (TASB). The PGM creates perturbation energy fields from multi-level features and extracts gradient-following feature trajectories that reflect the directionality of local responses. These trajectories are then processed by the TASB, a Mamba-based state-space unit that models dynamic propagation along each trajectory while incorporating velocity-constrained diffusion and semantically aligned feature fusion.
Unlike existing attention-based methods, TAPM-Net enables anisotropic, context-sensitive state transitions along spatial trajectories while maintaining global coherence at a low computational cost. The researchers tested TAPM-Net on two datasets, NUAA-SIRST and IRSTD-1K, and found that it achieved state-of-the-art performance in ISTD. This advancement could significantly enhance the accuracy and efficiency of infrared detection systems used in various industries, including energy, where thermal imaging is often employed for monitoring and maintenance purposes.
The research was published in the IEEE Transactions on Pattern Analysis and Machine Intelligence, a prestigious journal known for its rigorous peer-review process and high standards. This publication underscores the significance and potential impact of the researchers’ work in the field of computer vision and machine learning.
This article is based on research available at arXiv.