Scholarlink

Effective field management relies on proficient monitoring and control, as weeds compete with crops for essential resources such as moisture, nutrients, and sunlight. Integrating unmanned aerial vehicles (UAVs) with advanced weed detection and segmentation techniques enhances precision agriculture, enabling targeted weed management control. This study investigates the use of deep learning models to detect and segment weedy rice in rice fields utilizing images captured by UAVs. We employed various architectures, including YOLO (YOLOv8 and YOLOv11) and Mask R-CNN (implemented with the Detectron2 framework). The dataset consisted of 995 images, randomly split into training, validation, and test sets for evaluating models. Performance was evaluated using key metrics, including mean average precision (mAP50 and mAP50-90), precision, and recall. Among the models employed, YOLOv8s attained the highest performance, with a mAP50 of 0.843, a precision of 0.817, and a recall of 0.779, surpassing the other architectures. This study enhances computer vision applications in precision agriculture by demonstrating the effectiveness of deep learning models in detecting weedy rice using UAV imagery.

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

Weedy Rice Detection and Segmentation in UAV Imagery Using Deep Learning Models

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Van-Hoa Nguyen,
  • Cong-Doan Le,
  • Minh-Tuyen Truong,
  • Phuoc-Hai Huynh

摘要

Effective field management relies on proficient monitoring and control, as weeds compete with crops for essential resources such as moisture, nutrients, and sunlight. Integrating unmanned aerial vehicles (UAVs) with advanced weed detection and segmentation techniques enhances precision agriculture, enabling targeted weed management control. This study investigates the use of deep learning models to detect and segment weedy rice in rice fields utilizing images captured by UAVs. We employed various architectures, including YOLO (YOLOv8 and YOLOv11) and Mask R-CNN (implemented with the Detectron2 framework). The dataset consisted of 995 images, randomly split into training, validation, and test sets for evaluating models. Performance was evaluated using key metrics, including mean average precision (mAP50 and mAP50-90), precision, and recall. Among the models employed, YOLOv8s attained the highest performance, with a mAP50 of 0.843, a precision of 0.817, and a recall of 0.779, surpassing the other architectures. This study enhances computer vision applications in precision agriculture by demonstrating the effectiveness of deep learning models in detecting weedy rice using UAV imagery.