Scholarlink

<p>Accurate classification of papaya leaf diseases is essential for early intervention and yield protection. This study presents a&#xa0;comparative evaluation of four pre-trained convolutional neural networks—EfficientNetB3, InceptionV3, MobileNetV3, and VGG16—for classifying papaya foliar diseases, based on field-collected image data. The dataset included 1574 images categorized into Anthracnose, Bacterial Spot, Ring Spot, and Healthy classes. All models were trained and tested under identical settings, ensuring a&#xa0;fair comparison. EfficientNetB3 achieved a&#xa0;test accuracy of 94.58%, while VGG16 performed slightly better with 95.42%. InceptionV3 and MobileNetV3 followed with 92.50% and 92.92%, respectively. On a&#xa0;per-class basis, VGG16 maintained stable F1-scores, reaching 0.9753 for Ring Spot and 0.9550 for Anthracnose. In contrast, InceptionV3 showed greater variation, especially on Anthracnose with an F1-score of 0.9126. Weighted averages confirmed VGG16’s consistent performance across all classes. MobileNetV3, although slightly behind in accuracy, demonstrated efficiency suitable for lightweight deployment. These findings highlight that VGG16, under equal training conditions, provides stronger generalization for papaya leaf disease detection. The outcome is expected to assist researchers and practitioners in selecting reliable models for automated disease monitoring in precision agriculture.</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

Papaya Leaf Disease Classification Using Pre-trained Deep Learning Models: A Comparative Study

  • 期刊论文

  • 发表日期：

    2025年7月31日

查看全文
  • Yonis Gulzar

摘要

Accurate classification of papaya leaf diseases is essential for early intervention and yield protection. This study presents a comparative evaluation of four pre-trained convolutional neural networks—EfficientNetB3, InceptionV3, MobileNetV3, and VGG16—for classifying papaya foliar diseases, based on field-collected image data. The dataset included 1574 images categorized into Anthracnose, Bacterial Spot, Ring Spot, and Healthy classes. All models were trained and tested under identical settings, ensuring a fair comparison. EfficientNetB3 achieved a test accuracy of 94.58%, while VGG16 performed slightly better with 95.42%. InceptionV3 and MobileNetV3 followed with 92.50% and 92.92%, respectively. On a per-class basis, VGG16 maintained stable F1-scores, reaching 0.9753 for Ring Spot and 0.9550 for Anthracnose. In contrast, InceptionV3 showed greater variation, especially on Anthracnose with an F1-score of 0.9126. Weighted averages confirmed VGG16’s consistent performance across all classes. MobileNetV3, although slightly behind in accuracy, demonstrated efficiency suitable for lightweight deployment. These findings highlight that VGG16, under equal training conditions, provides stronger generalization for papaya leaf disease detection. The outcome is expected to assist researchers and practitioners in selecting reliable models for automated disease monitoring in precision agriculture.