Scholarlink

Mango foliage decay conditions hamper agronomical outputs consistently, resulting in substantial harvest losses and financial losses. Conventional leaf illness detection techniques are typically imprecise and involve a great deal of physical effort. We present a real-time solution using multiple deep learning approaches, incorporating both supervised and semi-supervised models to effectively detect mango leaf diseases and address these challenges. Remarkably, the best pre-trained model, DenseNet201, integrated with the semi-supervised FixMatch algorithm, achieved an impressive 99.93% accuracy using only 30% labeled data. With 500 photos per class, the dataset is divided into eight classes that correspond to different states of mango leaf. By merging pre-trained models with semi-supervised learning approaches, this innovative methodology effectively addresses the problem of real-time disease detection in agriculture in a scalable manner. Transparency is further improved via Explainable AI including interactive application, which offers insights into model decision-making mechanisms.

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

Not So Labeled Approach: FixMatch Outperforms Supervised Learning in Mango Leaf Disease Detection with XAI Insights

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Maksura Binte Rabbani Nuha,
  • Kazi Isat Mahazabin,
  • Md Tahsin,
  • Iffat Tasnim,
  • Nurzahan Akter Munni,
  • Al Hossain

摘要

Mango foliage decay conditions hamper agronomical outputs consistently, resulting in substantial harvest losses and financial losses. Conventional leaf illness detection techniques are typically imprecise and involve a great deal of physical effort. We present a real-time solution using multiple deep learning approaches, incorporating both supervised and semi-supervised models to effectively detect mango leaf diseases and address these challenges. Remarkably, the best pre-trained model, DenseNet201, integrated with the semi-supervised FixMatch algorithm, achieved an impressive 99.93% accuracy using only 30% labeled data. With 500 photos per class, the dataset is divided into eight classes that correspond to different states of mango leaf. By merging pre-trained models with semi-supervised learning approaches, this innovative methodology effectively addresses the problem of real-time disease detection in agriculture in a scalable manner. Transparency is further improved via Explainable AI including interactive application, which offers insights into model decision-making mechanisms.