Scholarlink

<p>To ensure that the agricultural sector increases productivity, quality, and sustainability, early detection of plant diseases is critical. Considerable effort has been made with deep learning and image processing for certain fruits including apples, citrus fruits, pomegranates, mangoes, and even papayas. Unfortunately, most of the models focus on a&#xa0;single fruit or a&#xa0;single image data type. In this work, we propose a&#xa0;multispecies deep learning framework for the detection and classification of diseases using explainable artificial intelligence (AI) with the integration of RGB (red–green–blue), thermal, and hyperspectral images. The methodology employs convolutional neural networks, YOLO (You Only Look Once), and some hybrid models for other parts of the methodology. More advanced learning methods for feature relevance pruning optimize these models. The system justifies its decisions using Grad-CAM (gradient-weighted class activation mapping) and SHAP (<i>Sh</i>apley <i>a</i>dditive ex<i>p</i>lanations), thus enhancing expert agricultural decision support, trust, and transparency. Addressing the problem of inaccurate and overfit models for new field data in the domain of spatial data, we used the tiered ontological structures that explain the hierarchical dependencies of domain knowledge. The inference time was less than 60 ms for each image, thus making it possible to use the system with mobile and Internet of Things (IoT) devices. Trust in dual explainability was more than 30% higher in user pilot studies. Such findings indicate that multimodal explainable fusion network (MEFN) technology can usher in the era of smart, sustainable, and scalable agriculture, as it allows for more robustness and generalization of fruits that are not visible as well as for easier, transparent, and user-friendly deployment. This research on adaptability, transparency, and diagnosis—features that make the solutions applicable to a&#xa0;range of fruit species and diseases—is an open problem area in fruit pathology AI and the smart-agriculture sector, to which it now contributes.</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

Explainable AI-Based Early Detection of Fruit Diseases Using a Multimodal Image Fusion Cross-Species Deep Learning Framework

  • 期刊论文

  • 发表日期：

    2025年9月26日

查看全文
  • Shruti Aggarwal,
  • Anil Kumar Verma

摘要

To ensure that the agricultural sector increases productivity, quality, and sustainability, early detection of plant diseases is critical. Considerable effort has been made with deep learning and image processing for certain fruits including apples, citrus fruits, pomegranates, mangoes, and even papayas. Unfortunately, most of the models focus on a single fruit or a single image data type. In this work, we propose a multispecies deep learning framework for the detection and classification of diseases using explainable artificial intelligence (AI) with the integration of RGB (red–green–blue), thermal, and hyperspectral images. The methodology employs convolutional neural networks, YOLO (You Only Look Once), and some hybrid models for other parts of the methodology. More advanced learning methods for feature relevance pruning optimize these models. The system justifies its decisions using Grad-CAM (gradient-weighted class activation mapping) and SHAP (Shapley additive explanations), thus enhancing expert agricultural decision support, trust, and transparency. Addressing the problem of inaccurate and overfit models for new field data in the domain of spatial data, we used the tiered ontological structures that explain the hierarchical dependencies of domain knowledge. The inference time was less than 60 ms for each image, thus making it possible to use the system with mobile and Internet of Things (IoT) devices. Trust in dual explainability was more than 30% higher in user pilot studies. Such findings indicate that multimodal explainable fusion network (MEFN) technology can usher in the era of smart, sustainable, and scalable agriculture, as it allows for more robustness and generalization of fruits that are not visible as well as for easier, transparent, and user-friendly deployment. This research on adaptability, transparency, and diagnosis—features that make the solutions applicable to a range of fruit species and diseases—is an open problem area in fruit pathology AI and the smart-agriculture sector, to which it now contributes.