Scholarlink

<p>Tomato maturity detection is an important foundation for intelligent harvesting and precision agriculture management. However, lighting variations, object occlusion, and background interference in complex farmland environments pose challenges to accurate recognition. To address these challenges, this paper proposes YOLO-TMD (Tomato Maturity Detection), an enhanced multi-modal detection model based on YOLOv11, integrating five key innovations to improve detection capability: the Cross-Modality Fusion (CMF) module achieves semantic alignment and dynamic fusion of RGB and Depth features; the C3k2-LGFC module enhances RGB texture and semantic representation; the C3k2-EPDC module strengthens Depth geometric structure and edge feature modeling; the Adaptive Fusion Dynamic Sampling (AFDS) module improves multi-scale feature reconstruction; and the Competitive Frequency Attention (CFA) module jointly enhances key features in both the frequency and spatial domains. Experimental results on the RGB-D tomato dataset show that YOLO-TMD achieves a Precision of 92.64%, Recall of 92.82%, mAP50 of 97.42%, and mAP50:95 of 82.62%, while maintaining a detection speed of 64.80 FPS, improving both detection accuracy and robustness compared to the baseline model. Ablation studies further validate the effectiveness of each module design and the stability of the model. The results indicate that YOLO-TMD can recognize tomatoes at different maturity stages in complex farmland environments, providing an efficient and robust technical solution for intelligent harvesting systems.</p> Graphical Abstract <p></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

YOLO-TMD: an RGB-D multi-modal network for tomato maturity detection

  • 期刊论文

  • 发表日期：

    2026年7月3日

查看全文
  • Yongqi Gai,
  • Jianguang Zhu,
  • Ruifan Wu

摘要

Tomato maturity detection is an important foundation for intelligent harvesting and precision agriculture management. However, lighting variations, object occlusion, and background interference in complex farmland environments pose challenges to accurate recognition. To address these challenges, this paper proposes YOLO-TMD (Tomato Maturity Detection), an enhanced multi-modal detection model based on YOLOv11, integrating five key innovations to improve detection capability: the Cross-Modality Fusion (CMF) module achieves semantic alignment and dynamic fusion of RGB and Depth features; the C3k2-LGFC module enhances RGB texture and semantic representation; the C3k2-EPDC module strengthens Depth geometric structure and edge feature modeling; the Adaptive Fusion Dynamic Sampling (AFDS) module improves multi-scale feature reconstruction; and the Competitive Frequency Attention (CFA) module jointly enhances key features in both the frequency and spatial domains. Experimental results on the RGB-D tomato dataset show that YOLO-TMD achieves a Precision of 92.64%, Recall of 92.82%, mAP50 of 97.42%, and mAP50:95 of 82.62%, while maintaining a detection speed of 64.80 FPS, improving both detection accuracy and robustness compared to the baseline model. Ablation studies further validate the effectiveness of each module design and the stability of the model. The results indicate that YOLO-TMD can recognize tomatoes at different maturity stages in complex farmland environments, providing an efficient and robust technical solution for intelligent harvesting systems.

Graphical Abstract