Scholarlink

Crop diseases are one of the biggest threats affecting agricultural production in the country, which has direct and indirect impacts affecting farmers and resulting in major economic losses. Numerous plant species, including rice, tomato, potato, tea, wheat, corn, and betel leaf, exhibit disease symptoms primarily on their leaves. The timely identification of visible disease patterns holds the potential to accurately forecast the onset of disease and facilitate proactive measures to mitigate its occurrence. Technological adoption and research advancements in the agricultural sector utilizing artificial intelligence (AI) have recently been observed. In this work, we explore the applicability of transfer learning methods with enhanced datasets for identification of crop diseases. For our experiment, we consider rice, corn, wheat, and potato, four major crops of Bangladesh and other Asian countries. Using images of infected and healthy plants, we evaluated three pre-trained models, DenseNet121, InceptionV3, and ResNet50. The best test accuracy for corn, potato, and rice was achieved by DenseNet121 among the models, with accuracies of 98.18%, 99.54%, and 91.13%, respectively. ResNet50 reached a flawless accuracy of 100% in detecting wheat disease. These results indicate the potential of transfer learning techniques in agricultural disease diagnosis.

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

Leveraging the Transfer Learning Techniques with Data Augmentation to Improve Bangladeshi Crop Diseases Diagnosis

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Ahmad Jainul Abidin,
  • Abdullah Al Maruf,
  • Md Kamrul Siam,
  • Rownuk Ara Rumy,
  • Zeyar Aung

摘要

Crop diseases are one of the biggest threats affecting agricultural production in the country, which has direct and indirect impacts affecting farmers and resulting in major economic losses. Numerous plant species, including rice, tomato, potato, tea, wheat, corn, and betel leaf, exhibit disease symptoms primarily on their leaves. The timely identification of visible disease patterns holds the potential to accurately forecast the onset of disease and facilitate proactive measures to mitigate its occurrence. Technological adoption and research advancements in the agricultural sector utilizing artificial intelligence (AI) have recently been observed. In this work, we explore the applicability of transfer learning methods with enhanced datasets for identification of crop diseases. For our experiment, we consider rice, corn, wheat, and potato, four major crops of Bangladesh and other Asian countries. Using images of infected and healthy plants, we evaluated three pre-trained models, DenseNet121, InceptionV3, and ResNet50. The best test accuracy for corn, potato, and rice was achieved by DenseNet121 among the models, with accuracies of 98.18%, 99.54%, and 91.13%, respectively. ResNet50 reached a flawless accuracy of 100% in detecting wheat disease. These results indicate the potential of transfer learning techniques in agricultural disease diagnosis.