Scholarlink

Objective <p><i>Erigeron annuus</i> and <i>Erigeron philadelphicus</i> are two common yet visually similar flowers, making it a challenging subject for human-observer identification but an ideal case for deep learning. Therefore, we aimed to establish an easily accessible image dataset to help beginners in deep learning distinguish these two flowers by using different models.</p> Results <p>We compiled “Fleabane”, a publicly available dataset comprising 400 floral images of the two species and validated its effectiveness for deep learning applications using 5-fold cross-validation in image classification and object detection after simple hyperparameter adjustment. For image classification using Inception V3, our model achieved a precision of 0.913, a recall of 0.978, an accuracy of 0.921, a F<sub>1</sub> score of 0.938, and an AUC of 0.987. For object detection with YOLOv8, it attained an mAP<sub>50</sub> of 0.977 and an mAP<sub>50−95</sub> of 0.675. These results show that deep learning, implemented using widely adopted frameworks, can effectively distinguish these species based on their floral appearance. The dataset is intentionally kept small and manageable, providing a practical and valuable resource for researchers, particularly those new to the field.</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

Exploring deep learning and data requirements through image classification of Erigeron annuus and Erigeron philadelphicus

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月16日

查看全文
  • Hiroshi Yamanaka,
  • Kohji Okamura

摘要

Objective

Erigeron annuus and Erigeron philadelphicus are two common yet visually similar flowers, making it a challenging subject for human-observer identification but an ideal case for deep learning. Therefore, we aimed to establish an easily accessible image dataset to help beginners in deep learning distinguish these two flowers by using different models.

Results

We compiled “Fleabane”, a publicly available dataset comprising 400 floral images of the two species and validated its effectiveness for deep learning applications using 5-fold cross-validation in image classification and object detection after simple hyperparameter adjustment. For image classification using Inception V3, our model achieved a precision of 0.913, a recall of 0.978, an accuracy of 0.921, a F1 score of 0.938, and an AUC of 0.987. For object detection with YOLOv8, it attained an mAP50 of 0.977 and an mAP50−95 of 0.675. These results show that deep learning, implemented using widely adopted frameworks, can effectively distinguish these species based on their floral appearance. The dataset is intentionally kept small and manageable, providing a practical and valuable resource for researchers, particularly those new to the field.