Scholarlink

We propose a computer vision pipeline to monitor coffee plant health from field-captured images. The methodology integrates three stages: (i) leaf detection with YOLOv8, (ii) optional contour segmentation with Segment Anything (SAM), and (iii) binary classification of each leaf (healthy vs. affected) using a ResNet-18 classifier fine-tuned on farm data. The system produces annotated visual outputs and per-plant metrics that facilitate auditing and early decision-making. On our self-collected full-plant image dataset, the pipeline achieves competitive results at each stage and yields a per-plant affected percentage consistent with qualitative inspection in the field. Beyond accuracy, we emphasize reproducibility (code/configs), calibrated decision thresholds, and deployment-oriented design.

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

Detection, Segmentation and Health Classification of Coffee Leaves in the Wild

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • César Osimani,
  • William Ruiz Martínez

摘要

We propose a computer vision pipeline to monitor coffee plant health from field-captured images. The methodology integrates three stages: (i) leaf detection with YOLOv8, (ii) optional contour segmentation with Segment Anything (SAM), and (iii) binary classification of each leaf (healthy vs. affected) using a ResNet-18 classifier fine-tuned on farm data. The system produces annotated visual outputs and per-plant metrics that facilitate auditing and early decision-making. On our self-collected full-plant image dataset, the pipeline achieves competitive results at each stage and yields a per-plant affected percentage consistent with qualitative inspection in the field. Beyond accuracy, we emphasize reproducibility (code/configs), calibrated decision thresholds, and deployment-oriented design.