Scholarlink

<p>Salient object detection (SOD) in optical remote sensing images (ORSIs) aims to identify and localize prominent targets amidst complex backgrounds. Challenges such as low contrast, shadow interference, and visually similar non-salient objects hinder SOD performance. To address these, we propose CEHAFNet, a contrast-enhanced hierarchical adaptive fusion network. CEHAFNet integrates a contrast enhancement module that utilizes contrast information to suppress encoder noise and enhance discriminative features, a channel relationship decoding block to model dependencies among high-level feature channels for distinguishing salient targets from visually similar distractors, and a hierarchical adaptive fusion module for global feature interaction. An adaptive multi-level loss function dynamically harmonizes supervision across pixel, region, and map levels. Experiments on ORSSD and EORSSD datasets demonstrate CEHAFNet’s leading performance among 16 state-of-the-art methods, with improvements in <InlineEquation ID="IEq1"> <EquationSource Format="TEX">\(S_{\alpha }\)</EquationSource> <EquationSource Format="MATHML"><math> <msub> <mi>S</mi> <mi>α</mi> </msub> </math></EquationSource> </InlineEquation> (0.9372 vs. 0.9003), <i>MAE</i> reduction by 12.38% (0.0092 vs. 0.0105), and <InlineEquation ID="IEq2"> <EquationSource Format="TEX">\(F_{\beta }^{max}\)</EquationSource> <EquationSource Format="MATHML"><math> <msubsup> <mi>F</mi> <mrow> <mi>β</mi> </mrow> <mrow> <mi mathvariant="italic">max</mi> </mrow> </msubsup> </math></EquationSource> </InlineEquation> enhancement by 0.0182 (from 0.9042 to 0.9224). The code is available at <a href="https://github.com/lsa2342/CEHAFNet">https://github.com/lsa2342/CEHAFNet</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

Enhanced saliency detection in remote sensing: a hierarchical adaptive fusion approach

  • 期刊论文

  • 发表日期：

    2026年6月24日

查看全文
  • Junwei Li,
  • Shuaiao Li,
  • Rihai Lai,
  • Peng Yu,
  • Xuefeng Ma

摘要

Salient object detection (SOD) in optical remote sensing images (ORSIs) aims to identify and localize prominent targets amidst complex backgrounds. Challenges such as low contrast, shadow interference, and visually similar non-salient objects hinder SOD performance. To address these, we propose CEHAFNet, a contrast-enhanced hierarchical adaptive fusion network. CEHAFNet integrates a contrast enhancement module that utilizes contrast information to suppress encoder noise and enhance discriminative features, a channel relationship decoding block to model dependencies among high-level feature channels for distinguishing salient targets from visually similar distractors, and a hierarchical adaptive fusion module for global feature interaction. An adaptive multi-level loss function dynamically harmonizes supervision across pixel, region, and map levels. Experiments on ORSSD and EORSSD datasets demonstrate CEHAFNet’s leading performance among 16 state-of-the-art methods, with improvements in \(S_{\alpha }\) S α (0.9372 vs. 0.9003), MAE reduction by 12.38% (0.0092 vs. 0.0105), and \(F_{\beta }^{max}\) F β max enhancement by 0.0182 (from 0.9042 to 0.9224). The code is available at https://github.com/lsa2342/CEHAFNet.