Scholarlink

<p>Infrared small target detection (IRSTD) remains a challenging task due to complex backgrounds, low signal-to-noise ratios, and the weak, small-scale nature of target features. Although existing methods have achieved promising results, they often face limitations in effectively distinguishing small targets from background clutter and maintaining robustness under varying environmental conditions. To address these challenges, this paper presents a novel IRSTD model based on an encoder-decoder architecture, specifically designed to enhance small target detection in complex infrared scenes. The encoder incorporates the Res2Net module combined with the convolutional block attention module (CBAM) to improve multi-scale feature extraction. In the decoder, we introduce residual convolution blocks (RCBs) and dilation bottleneck blocks (DBBs) to effectively capture global context and refine target localization. In addition, a combined loss function is proposed to improve detection accuracy while minimizing false alarm rate. Extensive experiments conducted on the SIRST, NUDT-SIRST and IRSTD-1K datasets demonstrate that our model outperforms state-of-the-art methods in both detection accuracy and false alarm reduction. Ablation studies validate the effectiveness of the proposed modules and show their impact on overall performance. Moreover, the model’s ability to maintain high processing speed using only CPU resources highlights its practicality for near real-time applications. The code is available at: <a href="https://github.com/long-nguyen12/irstd-pytorch">https://github.com/long-nguyen12/irstd-pytorch</a>.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Enhancing Infrared Small Target Detection: A Multi-scale Feature Integration Approach with an Optimized Combined Loss Function

  • 期刊论文

  • 发表日期：

    2026年2月9日

查看全文
  • Dinh Cong Nguyen,
  • Hoang Long Nguyen

摘要

Infrared small target detection (IRSTD) remains a challenging task due to complex backgrounds, low signal-to-noise ratios, and the weak, small-scale nature of target features. Although existing methods have achieved promising results, they often face limitations in effectively distinguishing small targets from background clutter and maintaining robustness under varying environmental conditions. To address these challenges, this paper presents a novel IRSTD model based on an encoder-decoder architecture, specifically designed to enhance small target detection in complex infrared scenes. The encoder incorporates the Res2Net module combined with the convolutional block attention module (CBAM) to improve multi-scale feature extraction. In the decoder, we introduce residual convolution blocks (RCBs) and dilation bottleneck blocks (DBBs) to effectively capture global context and refine target localization. In addition, a combined loss function is proposed to improve detection accuracy while minimizing false alarm rate. Extensive experiments conducted on the SIRST, NUDT-SIRST and IRSTD-1K datasets demonstrate that our model outperforms state-of-the-art methods in both detection accuracy and false alarm reduction. Ablation studies validate the effectiveness of the proposed modules and show their impact on overall performance. Moreover, the model’s ability to maintain high processing speed using only CPU resources highlights its practicality for near real-time applications. The code is available at: https://github.com/long-nguyen12/irstd-pytorch.