Scholarlink

<p>Object re-identification is a fundamental task in computer vision with wide applications in intelligent transportation and maritime security. In vessel re-identification, however, rainy and foggy conditions often cause severe image degradation, resulting in blurred contours, feature shifts, and identity confusion among vessels with similar appearance colors, which significantly degrades recognition performance. To address these challenges, this paper proposes a background structure-driven joint optimization framework, termed joint structure-prior learning (JSPL). JSPL introduces the residue channel prior (RCP) to model target-background structural relations at the channel level, effectively suppressing rain-fog interference while reducing reliance on color cues. In addition, a background structure enhancement module (BSEM) is designed to strengthen the representation of key contour regions, improving robustness to mixed degradations. Experimental results on <span>VesselReID_Adverse</span> dataset demonstrate that JSPL achieves superior feature extraction capability and environmental adaptability compared with mainstream methods. Ablation studies further verify the effectiveness of the background structure-guided module (BSGM) and BSEM, whose synergistic design enhances the model’s focus on structural identity features.</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

Vessel re-identification via background structure learning

  • 期刊论文

  • 发表日期：

    2026年5月11日

查看全文
  • Chi Xu,
  • Minxuan Yang,
  • Yiwen Wang,
  • Zhenzhu Fu,
  • Xian Zhong

摘要

Object re-identification is a fundamental task in computer vision with wide applications in intelligent transportation and maritime security. In vessel re-identification, however, rainy and foggy conditions often cause severe image degradation, resulting in blurred contours, feature shifts, and identity confusion among vessels with similar appearance colors, which significantly degrades recognition performance. To address these challenges, this paper proposes a background structure-driven joint optimization framework, termed joint structure-prior learning (JSPL). JSPL introduces the residue channel prior (RCP) to model target-background structural relations at the channel level, effectively suppressing rain-fog interference while reducing reliance on color cues. In addition, a background structure enhancement module (BSEM) is designed to strengthen the representation of key contour regions, improving robustness to mixed degradations. Experimental results on VesselReID_Adverse dataset demonstrate that JSPL achieves superior feature extraction capability and environmental adaptability compared with mainstream methods. Ablation studies further verify the effectiveness of the background structure-guided module (BSGM) and BSEM, whose synergistic design enhances the model’s focus on structural identity features.