Scholarlink

<p>In the era of artificial intelligence, plant disease prediction has gained the attention of researchers. This study explores the technologies of artificial intelligence for early disease prediction in rice plants. Diseases often cause subtle changes in the leaves before visual symptoms appear. Thermal imaging can capture these early signs. The proposed model leverages deep learning with two pre-trained algorithms (VGG16 and ResNet50) to improve prediction accuracy and reduce errors. The model focuses on Bacterial Leaf Blight (BLB) of the rice plant caused by Xanthomonas oryzae. Visual and thermal images of rice leaves were collected at three stages: healthy or normal (before infection), pre-symptomatic (48&#xa0;h after infection, with no visible symptoms yet), and post-symptomatic (after visible symptoms appear). The model compared normal leaves to pre-symptomatic leaves, achieving 97.56% accuracy for thermal images and 52.44% for visual images. Similar comparisons between normal and post-symptomatic leaves yielded high accuracy for both thermal (98.78%) and visual images (98.78%). To validate the model’s effectiveness, we compared it to various combinations of other deep learning models (VGG16, ResNet50, Xception, and InceptionV3). The proposed ensemble approach using VGG16 and ResNet50 achieved highest accuracy on the validation dataset of thermal images at the pre-symptomatic disease stage.</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

An Ensemble of Deep Transfer Learning Models for Plant Disease Prediction Using Thermal Images

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Ishita Bhakta,
  • Santanu Phadikar,
  • Koushik Majumder,
  • Arkaprabha Sau

摘要

In the era of artificial intelligence, plant disease prediction has gained the attention of researchers. This study explores the technologies of artificial intelligence for early disease prediction in rice plants. Diseases often cause subtle changes in the leaves before visual symptoms appear. Thermal imaging can capture these early signs. The proposed model leverages deep learning with two pre-trained algorithms (VGG16 and ResNet50) to improve prediction accuracy and reduce errors. The model focuses on Bacterial Leaf Blight (BLB) of the rice plant caused by Xanthomonas oryzae. Visual and thermal images of rice leaves were collected at three stages: healthy or normal (before infection), pre-symptomatic (48 h after infection, with no visible symptoms yet), and post-symptomatic (after visible symptoms appear). The model compared normal leaves to pre-symptomatic leaves, achieving 97.56% accuracy for thermal images and 52.44% for visual images. Similar comparisons between normal and post-symptomatic leaves yielded high accuracy for both thermal (98.78%) and visual images (98.78%). To validate the model’s effectiveness, we compared it to various combinations of other deep learning models (VGG16, ResNet50, Xception, and InceptionV3). The proposed ensemble approach using VGG16 and ResNet50 achieved highest accuracy on the validation dataset of thermal images at the pre-symptomatic disease stage.