Scholarlink

<p>This study introduces a low-cost optical screening method for distinguishing Huanglongbing (HLB)-infected citrus leaves from healthy and zinc-deficient counterparts using four spectral bands: red, green, blue, and infrared (IR). Three classification scenarios were evaluated using Decision Tree, k-Nearest Neighbors (k-NN), and Random Forest algorithms. In the binary classification of healthy and HLB-infected leaves (Group 1), the red and green bands showed the highest test accuracies, with Random Forest and k-NN achieving up to 0.8333 accuracy and precision above 0.84. In the ternary classification involving healthy, HLB-infected, and zinc-deficient leaves (Group 2), model performance declined due to reflectance data similarity between disease and deficiency symptoms, though Random Forest maintained moderate accuracy (0.6875) in the red band. The most complex scenario (Group 3), distinguishing healthy, blotchy mottle, green island, and zinc-deficient leaves, showed significant classification challenges, particularly for zinc-deficient samples. Findings highlight the red spectral band as the most effective for HLB detection and validate Random Forest as a robust algorithm for multi-class citrus leaf classification. This work supports the development of field-deployable screening tools for early disease detection for average farmers.</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

Optical Screening for Early Detection of Huanglongbing Infection and Zinc Deficiency in Citrus Leaves

  • 期刊论文

  • 发表日期：

    2026年5月20日

查看全文
  • Ramji Gupta,
  • Ashis Kumar Das,
  • Sushmita Mena,
  • Saurav Bharadwaj

摘要

This study introduces a low-cost optical screening method for distinguishing Huanglongbing (HLB)-infected citrus leaves from healthy and zinc-deficient counterparts using four spectral bands: red, green, blue, and infrared (IR). Three classification scenarios were evaluated using Decision Tree, k-Nearest Neighbors (k-NN), and Random Forest algorithms. In the binary classification of healthy and HLB-infected leaves (Group 1), the red and green bands showed the highest test accuracies, with Random Forest and k-NN achieving up to 0.8333 accuracy and precision above 0.84. In the ternary classification involving healthy, HLB-infected, and zinc-deficient leaves (Group 2), model performance declined due to reflectance data similarity between disease and deficiency symptoms, though Random Forest maintained moderate accuracy (0.6875) in the red band. The most complex scenario (Group 3), distinguishing healthy, blotchy mottle, green island, and zinc-deficient leaves, showed significant classification challenges, particularly for zinc-deficient samples. Findings highlight the red spectral band as the most effective for HLB detection and validate Random Forest as a robust algorithm for multi-class citrus leaf classification. This work supports the development of field-deployable screening tools for early disease detection for average farmers.