Scholarlink

Recently, the identification and classification of illnesses from images of plant leaves have been an ongoing problem in agricultural research. There is a dearth of proper dataset to be able to use modern deep learning-based disease detection algorithms for plant disease detection. We may rely less on farmers to take precautions to ensure the safety of their crops if they can use image-processing techniques to identify plant illnesses. For better use of these advancements, mobile applications need to be developed, where farmers directly can upload the images and get desired feedback for their crop/plant. In this work, we have developed such a solution where we attempted to solve the dataset crisis by using cycleGAN and used the generated synthetic dataset for leaf disease detection. For classification, we used transfer learning model EfficientNetB3 and achieved classification accuracy 93.33%. Finally, these developments are implemented in a mobile application for easy and scalable use for farmers.

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

A Coherent Solution of Dataset Crisis Using GAN Based Transfer Learning Approach for Leaf Image Classification and Disease Prediction

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Monu Bhagat,
  • Sunil Kumar

摘要

Recently, the identification and classification of illnesses from images of plant leaves have been an ongoing problem in agricultural research. There is a dearth of proper dataset to be able to use modern deep learning-based disease detection algorithms for plant disease detection. We may rely less on farmers to take precautions to ensure the safety of their crops if they can use image-processing techniques to identify plant illnesses. For better use of these advancements, mobile applications need to be developed, where farmers directly can upload the images and get desired feedback for their crop/plant. In this work, we have developed such a solution where we attempted to solve the dataset crisis by using cycleGAN and used the generated synthetic dataset for leaf disease detection. For classification, we used transfer learning model EfficientNetB3 and achieved classification accuracy 93.33%. Finally, these developments are implemented in a mobile application for easy and scalable use for farmers.