Scholarlink

<p>Crop diseases remain a significant threat to agricultural productivity and fruit quality, particularly for high-value crops such as strawberries and grapes. Early and reliable detection of these diseases under real-world conditions is essential but remains challenging due to variations in environment, illumination, and imaging perspectives. Leveraging recent advances in deep learning and computer vision, this study presents a robust framework for the automated detection and classification of strawberry and grape diseases using convolutional neural network (CNN) models, i.e., VGG16, ResNet101v2, InceptionV3, and DenseNet121. Unlike many existing works that rely solely on controlled or publicly available datasets, we constructed two specialized datasets by combining field-captured images under diverse environmental conditions with online sources, thereby enhancing robustness and ecological validity. The strawberry dataset includes six disease classes, while the grape dataset encompasses seven classes, covering economically significant pathologies such as anthracnose, black rot, gray mold, powdery mildew, sour rot, and leaf scorch. Extensive experiments were conducted using state-of-the-art CNN architectures, including VGG16, ResNet101v2, InceptionV3, and DenseNet121. On strawberries, DenseNet121 and InceptionV3 achieved accuracies of 94% (training) and 95% (testing), respectively, while VGG16 delivered superior performance on grapes, achieving 95% (training) and 92% (testing). Beyond technical accuracy, the proposed models were explicitly designed for applicability in actual field conditions, ensuring that the system can be directly adapted for use by farmers and plant pathologists as a practical decision-support tool. The findings provide a foundation for scalable, automated, and field-ready disease detection systems, contributing to more sustainable, data-driven crop management practices.</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

Early detection of strawberry and grape diseases under real-world field conditions using deep learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月17日

查看全文
  • Sultan Zeb,
  • Haleem Farman,
  • Muhammad Shabir,
  • Momina Shaheen,
  • Moustafa M. Nasralla

摘要

Crop diseases remain a significant threat to agricultural productivity and fruit quality, particularly for high-value crops such as strawberries and grapes. Early and reliable detection of these diseases under real-world conditions is essential but remains challenging due to variations in environment, illumination, and imaging perspectives. Leveraging recent advances in deep learning and computer vision, this study presents a robust framework for the automated detection and classification of strawberry and grape diseases using convolutional neural network (CNN) models, i.e., VGG16, ResNet101v2, InceptionV3, and DenseNet121. Unlike many existing works that rely solely on controlled or publicly available datasets, we constructed two specialized datasets by combining field-captured images under diverse environmental conditions with online sources, thereby enhancing robustness and ecological validity. The strawberry dataset includes six disease classes, while the grape dataset encompasses seven classes, covering economically significant pathologies such as anthracnose, black rot, gray mold, powdery mildew, sour rot, and leaf scorch. Extensive experiments were conducted using state-of-the-art CNN architectures, including VGG16, ResNet101v2, InceptionV3, and DenseNet121. On strawberries, DenseNet121 and InceptionV3 achieved accuracies of 94% (training) and 95% (testing), respectively, while VGG16 delivered superior performance on grapes, achieving 95% (training) and 92% (testing). Beyond technical accuracy, the proposed models were explicitly designed for applicability in actual field conditions, ensuring that the system can be directly adapted for use by farmers and plant pathologists as a practical decision-support tool. The findings provide a foundation for scalable, automated, and field-ready disease detection systems, contributing to more sustainable, data-driven crop management practices.