DCCS-Det: Directional Context and Cross-Scale-Aware Detector for Infrared Small Target

TL;DR

DCCS-Det tackles infrared small target detection by explicitly modeling both local detail and long-range directional context through a Dual-stream Saliency Enhancement block and by preserving deep target semantics with Latent-aware Semantic Extraction and Aggregation. The dual-branch design enables joint feature modeling with an auxiliary encoder, while LaSEA provides cross-scale semantic guidance via random pooling regularization and cross-dilated feature extraction. Ablation and comparative analyses show the approach achieves state-of-the-art IoU on multiple IRSTD benchmarks with low false alarms and efficient inference, highlighting its practical impact for remote sensing and surveillance. The work suggests strong potential for real-world deployment and future exploration of lightweight architectures and multi-domain feature fusion.

Abstract

Infrared small target detection (IRSTD) is critical for applications like remote sensing and surveillance, which aims to identify small, low-contrast targets against complex backgrounds. However, existing methods often struggle with inadequate joint modeling of local-global features (harming target-background discrimination) or feature redundancy and semantic dilution (degrading target representation quality). To tackle these issues, we propose DCCS-Det (Directional Context and Cross-Scale Aware Detector for Infrared Small Target), a novel detector that incorporates a Dual-stream Saliency Enhancement (DSE) block and a Latent-aware Semantic Extraction and Aggregation (LaSEA) module. The DSE block integrates localized perception with direction-aware context aggregation to help capture long-range spatial dependencies and local details. On this basis, the LaSEA module mitigates feature degradation via cross-scale feature extraction and random pooling sampling strategies, enhancing discriminative features and suppressing noise. Extensive experiments show that DCCS-Det achieves state-of-the-art detection accuracy with competitive efficiency across multiple datasets. Ablation studies further validate the contributions of DSE and LaSEA in improving target perception and feature representation under complex scenarios. \href{https://huggingface.co/InPeerReview/InfraredSmallTargetDetection-IRSTD.DCCS}{DCCS-Det Official Code is Available Here!}

Figures (9)

• Figure 1: Complete framework of the DCCS-Det method. First, the infrared image is fed into the Dual-Branch Joint Extraction to model features. Then, the Structure Refinement enhances the representation of the target semantics in the deeper layers of the network to guide fusion in the decoder. Finally, the Progressive Supervision Head utilizes the fused features to predict the final result. In the figure, PD represents the predicted result, and GT represents the Ground Truth (actual label).
• Figure 2: This figure illustrates the architecture of the DSE Block. This module consists of two branches: the left branch enhances local detail features; the right branch employs 2D-Selective-Scan (SS2D) to capture global contextual information with linear complexity. These two feature streams join with the primary branch to achieve collaborative modeling of local details and global dependencies of the target.
• Figure 3: This figure illustrates the architecture of the LaSEA module. The module adopts cross-scale feature extraction and random pooling sampling strategies to provide more discriminative and robust target semantic feature representations, thereby enhancing the capability of guiding the decoder in fusion.
• Figure 4: Comparison of ROC curves across various approaches on (a) IRSTD-1K, (b) NUAA-SIRST, and (c) SIRST-Aug datasets. The horizontal axis represents FPR while the vertical axis represents TPR. Curves approaching the top-left corner indicate superior detection performance.
• Figure 5: Visual results of different IRSTD methods on the IRSTD-1K, NUAA-SIRST, and SIRST-Aug datasets. Blue, yellow, and red circles respectively denote correctly detected targets, missed targets, and false alarms.