Scholarlink

Guava (Psidium guajava), a key fruit crop, is susceptible to various diseases that threaten its productivity. This study presents a deep learning and computer vision to detect guava diseases in both leaves and fruits. A Convolutional Neural Network (CNN) was initially implemented, followed by data augmentation and transfer learning to enhance classification performance. Several pre-trained models, including EfficientNetB3, InceptionV3, Xception, VGG16, MobileNetV2, ResNet50V2, DenseNet121 and GoogleNet (Inception V1), were trained and evaluated on six disease categories: Anthracnose, Stylar-End-Rot, Scab, Red-Rust, Phytophthora and disease-free samples. To assess model performance accuracy, precision, recall and F1-score were calculated for every model. While multiple models achieved high accuracy (97–98%), GoogleNet (InceptionV1) outperformed others with 99.16% accuracy. The study’s findings demonstrate improvements over previous methods, highlighting the effectiveness of the proposed approach.

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

Deep Learning-Driven Detection of Guava Diseases for Smart Agriculture

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • M. Prasanna Kumari,
  • G. N. V. G. Sirisha,
  • R. Amith Varma

摘要

Guava (Psidium guajava), a key fruit crop, is susceptible to various diseases that threaten its productivity. This study presents a deep learning and computer vision to detect guava diseases in both leaves and fruits. A Convolutional Neural Network (CNN) was initially implemented, followed by data augmentation and transfer learning to enhance classification performance. Several pre-trained models, including EfficientNetB3, InceptionV3, Xception, VGG16, MobileNetV2, ResNet50V2, DenseNet121 and GoogleNet (Inception V1), were trained and evaluated on six disease categories: Anthracnose, Stylar-End-Rot, Scab, Red-Rust, Phytophthora and disease-free samples. To assess model performance accuracy, precision, recall and F1-score were calculated for every model. While multiple models achieved high accuracy (97–98%), GoogleNet (InceptionV1) outperformed others with 99.16% accuracy. The study’s findings demonstrate improvements over previous methods, highlighting the effectiveness of the proposed approach.