Scholarlink

<p>Accurate, non-destructive identification of citrus species through leaf-emitted volatile organic compounds (VOCs) is essential for modern precision agriculture, supporting cultivar authentication, orchard management, and conservation of region-specific biodiversity. This study introduces a novel, cost-effective electronic nose (E-Nose) system designed to distinguish seven citrus species native to North-eastern India. The system features an iteratively optimized sensor chamber, developed through finite element method (FEM) simulations, ensuring uniform VOC distribution. A processed dataset of 8,354 E-Nose sensor responses, acquired from 70 composite citrus leaf samples, was used to benchmark nine machine learning classifiers with rigorous hyperparameter tuning and cross-validation. The combination of Linear Discriminant Analysis (LDA) and Bagging k-Nearest Neighbors (k-NN) achieved 99.52% test accuracy and a Cohen’s Kappa of 99.44% on the independent held-out test set. Stratified cross-validation confirmed consistent performance within the same controlled experiment, and clear separation was observed even between chemically similar cultivars. Visual analyses confirmed robust classification with minimal misclassifications. Importantly, when evaluated using strict tree-level splitting (all measurements from the same tree kept together), the final classifier still achieved 96.58% accuracy and macro-AUC of 0.990 on completely unseen trees, confirming robust generalisation to new plants. By enabling rapid and reliable species identification, this E-Nose framework supports accurate cultivar labelling, aids in securing Protected Designation of Origin (PDO) status, and contributes to early disease surveillance and sustainable orchard practices. The study provides a validated baseline for scaling the system to field conditions, paving the way for practical, scalable E-Nose solutions in citrus agriculture.</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

Citrus cultivar identification using a computationally optimized electronic nose system and machine learning approach

  • 期刊论文

  • 发表日期：

    2026年1月17日

查看全文
  • Sudipta Hazarika,
  • Rajdeep Choudhury,
  • Babak Montazer,
  • Chandra Kamal Borah,
  • Arindam Phukan,
  • Lakhinath Borah,
  • Utpal Sarma

摘要

Accurate, non-destructive identification of citrus species through leaf-emitted volatile organic compounds (VOCs) is essential for modern precision agriculture, supporting cultivar authentication, orchard management, and conservation of region-specific biodiversity. This study introduces a novel, cost-effective electronic nose (E-Nose) system designed to distinguish seven citrus species native to North-eastern India. The system features an iteratively optimized sensor chamber, developed through finite element method (FEM) simulations, ensuring uniform VOC distribution. A processed dataset of 8,354 E-Nose sensor responses, acquired from 70 composite citrus leaf samples, was used to benchmark nine machine learning classifiers with rigorous hyperparameter tuning and cross-validation. The combination of Linear Discriminant Analysis (LDA) and Bagging k-Nearest Neighbors (k-NN) achieved 99.52% test accuracy and a Cohen’s Kappa of 99.44% on the independent held-out test set. Stratified cross-validation confirmed consistent performance within the same controlled experiment, and clear separation was observed even between chemically similar cultivars. Visual analyses confirmed robust classification with minimal misclassifications. Importantly, when evaluated using strict tree-level splitting (all measurements from the same tree kept together), the final classifier still achieved 96.58% accuracy and macro-AUC of 0.990 on completely unseen trees, confirming robust generalisation to new plants. By enabling rapid and reliable species identification, this E-Nose framework supports accurate cultivar labelling, aids in securing Protected Designation of Origin (PDO) status, and contributes to early disease surveillance and sustainable orchard practices. The study provides a validated baseline for scaling the system to field conditions, paving the way for practical, scalable E-Nose solutions in citrus agriculture.