Scholarlink

Plant diseases represent one of the important threats to the agricultural sector, causing crop damage and leading to catastrophic economic losses in most countries, among them Morocco. Therefore, the development of accurate and intelligent applications based on computer vision is essential, as they enable the detection of plant diseases and provide farmers with the opportunity to take timely decisions to prevent yield losses. For this purpose, in this paper, we propose a hybrid approach that combines the MobileNetV3-Small model based on CNN architecture and the Vision Transformer model to develop an intelligent mobile application that can be used by the farmers of the region of Settat, Morocco. The mobile application is designed to detect plant diseases at an early stage and support farmers in taking the necessary steps before crops suffer further damage. For the training phase, we used the PlantVillage Dataset, which contains 54,305 images across 38 classes, covering 14 crop species and 26 distinct diseases. The experimental tests show that the combination of MobileNetV3-Smal model and the Vision Transformer model achieved promissing performance, reaching 95.19% accuracy on the training dataset, 96.78% accuracy on the validation dataset, and 96.48% accuracy on the testing dataset. These results highlight the potential of this hybrid approach for developing an accurate and reliable mobile application for plant disease detection.

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

AgriVision Model: Combining Convolutional Neural Networks and Transformers for Plant Leaf Disease Detection

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sara Belattar,
  • Fatima-Zohra Hibbi,
  • Peter Rus

摘要

Plant diseases represent one of the important threats to the agricultural sector, causing crop damage and leading to catastrophic economic losses in most countries, among them Morocco. Therefore, the development of accurate and intelligent applications based on computer vision is essential, as they enable the detection of plant diseases and provide farmers with the opportunity to take timely decisions to prevent yield losses. For this purpose, in this paper, we propose a hybrid approach that combines the MobileNetV3-Small model based on CNN architecture and the Vision Transformer model to develop an intelligent mobile application that can be used by the farmers of the region of Settat, Morocco. The mobile application is designed to detect plant diseases at an early stage and support farmers in taking the necessary steps before crops suffer further damage. For the training phase, we used the PlantVillage Dataset, which contains 54,305 images across 38 classes, covering 14 crop species and 26 distinct diseases. The experimental tests show that the combination of MobileNetV3-Smal model and the Vision Transformer model achieved promissing performance, reaching 95.19% accuracy on the training dataset, 96.78% accuracy on the validation dataset, and 96.48% accuracy on the testing dataset. These results highlight the potential of this hybrid approach for developing an accurate and reliable mobile application for plant disease detection.