Scholarlink

<p>Leaf segmentation in real-world scenarios is particularly challenging due to varying lighting and background conditions, crowded environments, and overlapping foliage. Existing techniques and methodologies often underperform in these contexts due to their inherent constraints of relying only on RGB or thermal images. While RGB data is sensitive to variations in lighting, thermal data lacks the structural details required for accurate segmentation. This study introduces an adaptive attention-based deep learning framework that integrates RGB and temperature modalities to address these challenges. The fusion mechanism allows the model to leverage the complementary qualities of both inputs, improving its ability to capture detailed leaf properties. The proposed model is evaluated using a dedicated dataset of paired RGB and thermal images, with performance assessed across seven metrics: Accuracy, Jaccard Index, Precision, Recall, F1-score, mAP@0.5, mAP@0.75, and IoU. The results demonstrate that the method consistently outperforms current state-of-the-art techniques across all evaluation metrics.</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

An adaptive attention-based multimodal deep learning framework integrating RGB and thermal imaging for robust leaf segmentation in real-world environments

  • 期刊论文

  • 发表日期：

    2026年5月12日

查看全文
  • Ankur Kulhari,
  • Sanjeev Patwa

摘要

Leaf segmentation in real-world scenarios is particularly challenging due to varying lighting and background conditions, crowded environments, and overlapping foliage. Existing techniques and methodologies often underperform in these contexts due to their inherent constraints of relying only on RGB or thermal images. While RGB data is sensitive to variations in lighting, thermal data lacks the structural details required for accurate segmentation. This study introduces an adaptive attention-based deep learning framework that integrates RGB and temperature modalities to address these challenges. The fusion mechanism allows the model to leverage the complementary qualities of both inputs, improving its ability to capture detailed leaf properties. The proposed model is evaluated using a dedicated dataset of paired RGB and thermal images, with performance assessed across seven metrics: Accuracy, Jaccard Index, Precision, Recall, F1-score, mAP@0.5, mAP@0.75, and IoU. The results demonstrate that the method consistently outperforms current state-of-the-art techniques across all evaluation metrics.