Title: RPCASSM: A Robust PCA State Space Model for Detecting Infrared Small Targets

Abstract:

Accurate detection and segmentation of infrared small targets are critical for applications in maritime rescue, surveillance, and security. However, because these targets occupy a minimal area in long-distance imaging, mainstream visual state space models often struggle with efficiency and fail to precisely capture target boundaries. Current infrared-specific state space models have largely retained the structural framework of general visual models, neglecting the unique characteristics of infrared small targets.

To address these limitations, this study introduces RPCASSM, a network built upon the paradigm of Robust Principal Component Analysis (RPCA). This approach leverages the spatial domain properties of infrared small targets to construct two specialized components: the Background State Space Module (BSSM) and the Target State Space Module (TSSM). The BSSM employs a Spatial Probe Scanning Mechanism (SPCM) to model background information by exploiting the saliency of spatially heterogeneous signals. Meanwhile, the TSSM utilizes a Deformable Prompt Scanning Mechanism (DPCM), which capitalizes on the sparsity and local highlights of targets to perform state space modeling within the target’s deformable space.

This architectural design effectively resolves the challenge faced by existing mainstream vision state space models in accurately representing the edge structures of infrared small targets. Validations on standard benchmark datasets confirm the efficacy of the RPCASSM framework. The source code for this project will be publicly available at https://github.com/PepperCS/RPCASSM.

Source: arXiv Generated at: 2026-06-02 00:00:00 UTC