Scholarlink

<p>Attention U-Net to accurately segment turmeric leaf diseases with the help of attention gates to improve the extraction of discriminative features and inhibit undesirable background data. The turmeric leaf model is trained on a large dataset of turmeric leaves, with extensive data augmentation to enhance generalization in real-world environments. The suggested framework allows for the localization of diseased areas with high precision at the pixel level and calculates the percentage of diseased leaf area to determine the objective severity. Guidance of skip connections by attention can enhance boundary delineation of irregular lesions and increase robustness to illumination changes, leaf orientation variations, and background complexity; thus, the strategy is appropriate for precision agriculture-based disease monitoring and control. The performance is shown to be much higher, with an average Intersection over Union of 0.82, Dice Coefficient of 0.89, Pixel Accuracy of 0.96, precision of 0.90, recall of 0.89, and weighted loss of 0.24, and inference time of 55 ms per image, which is better than existing methods.</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

An attention U-Net–enabled precision agriculture pipeline for turmeric disease detection and mapping

  • 期刊论文

  • 发表日期：

    2026年3月4日

查看全文
  • V. Bhavani,
  • G. Pradeepini,
  • K. Ch Sri Kavya

摘要

Attention U-Net to accurately segment turmeric leaf diseases with the help of attention gates to improve the extraction of discriminative features and inhibit undesirable background data. The turmeric leaf model is trained on a large dataset of turmeric leaves, with extensive data augmentation to enhance generalization in real-world environments. The suggested framework allows for the localization of diseased areas with high precision at the pixel level and calculates the percentage of diseased leaf area to determine the objective severity. Guidance of skip connections by attention can enhance boundary delineation of irregular lesions and increase robustness to illumination changes, leaf orientation variations, and background complexity; thus, the strategy is appropriate for precision agriculture-based disease monitoring and control. The performance is shown to be much higher, with an average Intersection over Union of 0.82, Dice Coefficient of 0.89, Pixel Accuracy of 0.96, precision of 0.90, recall of 0.89, and weighted loss of 0.24, and inference time of 55 ms per image, which is better than existing methods.