Scholarlink

<p>Remote sensing (RS) images capture multispectral super-resolution features essential for applications like geological analysis, weather prediction, and urban planning. Traditional retrieval methods often yield suboptimal results due to the lack of learnable parameters. To address this, we propose a novel conflict learning approach within a generative model framework for RS image retrieval. The proposed method integrates an encoder-decoder semantic network, enhancing performance through continuous conflict learning between generator and discriminator components. This allows the system to extract the most critical multispectral features. The proposed approach surpasses state-of-the-art methods, achieving 0.9782 and 0.9974 mAP on UCMD and PatternNet datasets, respectively. Additionally, it enhances retrieval efficiency, reducing ANMRR to 0.0176 and 0.0029 for both datasets. These results demonstrate the effectiveness of our method in improving accuracy and efficiency in RS image retrieval.</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

A conflict-aware learning network in generative models for enhancing retrieval efficiency of remote sensing images

  • 期刊论文

  • 发表日期：

    2025年9月29日

查看全文
  • Salahuddin Unar,
  • Mobeen Nazar,
  • Farooque Hassan Kumbhar,
  • Salah-ud-din Khokhar,
  • Delanyo Kwame Bensah Kulevome,
  • Ahmed Elkhalil

摘要

Remote sensing (RS) images capture multispectral super-resolution features essential for applications like geological analysis, weather prediction, and urban planning. Traditional retrieval methods often yield suboptimal results due to the lack of learnable parameters. To address this, we propose a novel conflict learning approach within a generative model framework for RS image retrieval. The proposed method integrates an encoder-decoder semantic network, enhancing performance through continuous conflict learning between generator and discriminator components. This allows the system to extract the most critical multispectral features. The proposed approach surpasses state-of-the-art methods, achieving 0.9782 and 0.9974 mAP on UCMD and PatternNet datasets, respectively. Additionally, it enhances retrieval efficiency, reducing ANMRR to 0.0176 and 0.0029 for both datasets. These results demonstrate the effectiveness of our method in improving accuracy and efficiency in RS image retrieval.