Scholarlink

<p>Sugarcane is a&#xa0;crop of major strategic importance within global agro-industrial systems; however, its productivity and quality are increasingly threatened by a&#xa0;wide spectrum of foliar diseases. Early and accurate disease identification is therefore critical for maintaining crop health, optimizing agricultural input usage, and mitigating significant yield losses. This study provides a&#xa0;comprehensive analytical evaluation of a&#xa0;curated sugarcane leaf disease dataset and presents a&#xa0;systematic benchmarking of state-of-the-art deep learning architectures for automated disease classification in agricultural imagery. To ensure methodological rigor, reproducibility, and fair performance comparison, a&#xa0;structured experiment management framework was employed consistently throughout the experimental pipeline. Experimental findings demonstrate that EfficientNet-B3 achieves the highest overall performance, attaining accuracy, precision, recall, and F1-score of 98.94%, along with a&#xa0;near-perfect macro-ROC-AUC of 0.99876. ViT-B/16 and ResNet50 also exhibit strong classification performance, achieving accuracies of 96.82% and 96.29%, respectively. Beyond quantitative evaluation, attention-based interpretability analyses indicate that the evaluated models effectively focus on disease-relevant visual features, including lesion structures and discoloration regions. These findings confirm the models’ ability to learn biologically meaningful patterns associated with plant pathology, supporting their potential applicability in precision agriculture and sustainable sugarcane production.</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 Analytical Review of Deep Learning Models for Sugarcane Leaf Disease Classification

  • 期刊论文

  • 发表日期：

    2026年4月8日

查看全文
  • Zahraa Tarek,
  • Esraa Hassan

摘要

Sugarcane is a crop of major strategic importance within global agro-industrial systems; however, its productivity and quality are increasingly threatened by a wide spectrum of foliar diseases. Early and accurate disease identification is therefore critical for maintaining crop health, optimizing agricultural input usage, and mitigating significant yield losses. This study provides a comprehensive analytical evaluation of a curated sugarcane leaf disease dataset and presents a systematic benchmarking of state-of-the-art deep learning architectures for automated disease classification in agricultural imagery. To ensure methodological rigor, reproducibility, and fair performance comparison, a structured experiment management framework was employed consistently throughout the experimental pipeline. Experimental findings demonstrate that EfficientNet-B3 achieves the highest overall performance, attaining accuracy, precision, recall, and F1-score of 98.94%, along with a near-perfect macro-ROC-AUC of 0.99876. ViT-B/16 and ResNet50 also exhibit strong classification performance, achieving accuracies of 96.82% and 96.29%, respectively. Beyond quantitative evaluation, attention-based interpretability analyses indicate that the evaluated models effectively focus on disease-relevant visual features, including lesion structures and discoloration regions. These findings confirm the models’ ability to learn biologically meaningful patterns associated with plant pathology, supporting their potential applicability in precision agriculture and sustainable sugarcane production.