Scholarlink

<p>Sericulture is a vital agro-based sector in Asia, particularly in India, where it supports sustainable livelihoods for millions of rural households and makes a significant contribution to the country’s agricultural economy. The productivity and quality of silk production are strongly dependent on the health of the mulberry (<i>Morus alba</i>), the exclusive host plant of the silkworm (<i>Bombyx mori</i>). Foliar diseases such as rust (<i>Cercospora moricola</i>) and leaf spot (<i>Pseudocercospora mori</i>) pose serious threats to mulberry cultivation, causing up to 30–40% losses in leaf biomass and adversely affecting leaf nutritional quality, chlorophyll content, and silkworm performance. Early and accurate identification of these diseases is therefore essential for sustainable sericulture and effective crop management. This study proposes an artificial intelligence-enabled framework for the automated detection of mulberry leaf diseases using efficient deep learning models. A publicly available dataset comprising healthy, rust-infected, and leaf spot-infected mulberry leaves was curated, balanced, and augmented to form a dataset of 8,180 images. To improve disease feature representation and minimize background interference, leaf segmentation was performed using a U²-net–based background removal approach. Transfer learning was employed to fine-tune EfficientNetV2-S, EfficientNetV2-M, and EfficientNetV2-L models, with a focus on identifying architectures suitable for real-time and resource-constrained agricultural applications. Among the evaluated models, EfficientNetV2-S demonstrated the best trade-off between accuracy and computational efficiency, achieving a classification accuracy of 91.20% and an F1-score of 0.91, while requiring significantly lower computational resources than the larger variants. The results indicate that lightweight deep learning models can deliver reliable disease diagnosis without reliance on high-end hardware, making them well-suited for on-field and mobile-based deployment. The proposed framework offers a cost-effective, non-invasive, and scalable solution for precision disease management in mulberry cultivation. By enabling timely intervention and reducing yield losses, this approach supports sustainable sericulture practices, enhances rural livelihoods, and contributes to long-term agricultural sustainability.</p>

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

AI-driven mulberry leaf disease detection for sustainable and resilient sericulture systems

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月23日

查看全文
  • N. Bharath Kumar,
  • Krishna Kumar Chaturvedi,
  • Rajeev Ranjan Kumar,
  • L. Suman,
  • A. Praveen Kumar,
  • Shivadarshan Jirli,
  • Snehasis Mallik,
  • H. Shabreen Taj

摘要

Sericulture is a vital agro-based sector in Asia, particularly in India, where it supports sustainable livelihoods for millions of rural households and makes a significant contribution to the country’s agricultural economy. The productivity and quality of silk production are strongly dependent on the health of the mulberry (Morus alba), the exclusive host plant of the silkworm (Bombyx mori). Foliar diseases such as rust (Cercospora moricola) and leaf spot (Pseudocercospora mori) pose serious threats to mulberry cultivation, causing up to 30–40% losses in leaf biomass and adversely affecting leaf nutritional quality, chlorophyll content, and silkworm performance. Early and accurate identification of these diseases is therefore essential for sustainable sericulture and effective crop management. This study proposes an artificial intelligence-enabled framework for the automated detection of mulberry leaf diseases using efficient deep learning models. A publicly available dataset comprising healthy, rust-infected, and leaf spot-infected mulberry leaves was curated, balanced, and augmented to form a dataset of 8,180 images. To improve disease feature representation and minimize background interference, leaf segmentation was performed using a U²-net–based background removal approach. Transfer learning was employed to fine-tune EfficientNetV2-S, EfficientNetV2-M, and EfficientNetV2-L models, with a focus on identifying architectures suitable for real-time and resource-constrained agricultural applications. Among the evaluated models, EfficientNetV2-S demonstrated the best trade-off between accuracy and computational efficiency, achieving a classification accuracy of 91.20% and an F1-score of 0.91, while requiring significantly lower computational resources than the larger variants. The results indicate that lightweight deep learning models can deliver reliable disease diagnosis without reliance on high-end hardware, making them well-suited for on-field and mobile-based deployment. The proposed framework offers a cost-effective, non-invasive, and scalable solution for precision disease management in mulberry cultivation. By enabling timely intervention and reducing yield losses, this approach supports sustainable sericulture practices, enhances rural livelihoods, and contributes to long-term agricultural sustainability.