Scholarlink

Potato late blight is a devastating plant disease that significantly impacts agricultural productivity worldwide. This study evaluates the effectiveness of different data augmentation techniques in enhancing the performance of the YOLOv8m model for disease detection. The model was trained and tested using datasets augmented with brightness, contrast, Gaussian blur, and a final one that combined all of them. The results show that while individual augmentations to the original dataset slightly improved the model's performance, contrast manipulation substantially improved the model's precision, recall, and robustness, achieving a maximum mAP50 of 92.4% and a maximum F1 score of 86%. These results suggest that a well-tailored and optimized preprocessing pipeline greatly benefits the model's disease detection capabilities while reducing the computational load in incremental learning. This research sets the pathway for further developing and integrating these technologies to provide customized solutions for agricultural stakeholders.

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

Multi-level Sequential Data Augmentation Approach for Assessing Vision Models Performance in Agrarian Domain

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yassine Zarrouk,
  • Mohammed Bourhaleb,
  • Mohammed Rahmoune,
  • Khalid Hachami,
  • Mimoun Yandouzi

摘要

Potato late blight is a devastating plant disease that significantly impacts agricultural productivity worldwide. This study evaluates the effectiveness of different data augmentation techniques in enhancing the performance of the YOLOv8m model for disease detection. The model was trained and tested using datasets augmented with brightness, contrast, Gaussian blur, and a final one that combined all of them. The results show that while individual augmentations to the original dataset slightly improved the model's performance, contrast manipulation substantially improved the model's precision, recall, and robustness, achieving a maximum mAP50 of 92.4% and a maximum F1 score of 86%. These results suggest that a well-tailored and optimized preprocessing pipeline greatly benefits the model's disease detection capabilities while reducing the computational load in incremental learning. This research sets the pathway for further developing and integrating these technologies to provide customized solutions for agricultural stakeholders.