Scholarlink

<p>Black pepper, a cornerstone crop in the agricultural landscape of northern Kerala, India, continues to face severe challenges due to diseases and nutrient deficiencies. Timely identification of these issues is vital for sustaining productivity and minimizing economic losses. This study proposes an advanced deep learning framework comprising ensemble-based classification models for automated detection of black pepper leaf diseases and nutrient deficiencies. Leveraging pre-trained Convolutional Neural Networks (CNNs), six state-of-the-art architectures were fine-tuned and evaluated individually and in ensemble configurations. A custom dataset, the KSD-BPLDND Dataset, containing 8,469 field images across ten classes, was developed to facilitate model training and evaluation. To assess the generalization capability of the proposed system, we additionally validated the models on an external benchmark–the Potato Leaf Disease Dataset comprising 3,076 images across seven disease categories. The cross-dataset evaluation confirms the robustness and transferability of the ensemble models beyond the target crop. Experimental results demonstrate that the majority voting ensemble (BPLDNDNet-MV) achieves 99.53% accuracy on the KSD-BPLDND dataset and maintains high performance on the external potato dataset, reinforcing the general-purpose applicability of the approach. The early fusion ensemble (BPLDNDNet-EF) also achieved strong results with 98.99% accuracy. These findings highlight the potential of the proposed frameworks to support scalable, reliable plant health monitoring in diverse agricultural scenarios.</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

Ensemble-based Deep Feature Extraction using Pre-trained CNNs for Black Pepper Leaf Disease and Nutrient Deficiency Detection and Classification: BPLDNDNet

  • 期刊论文

  • 发表日期：

    2026年2月2日

查看全文
  • Ratheesh Raju,
  • T. M. Thasleema

摘要

Black pepper, a cornerstone crop in the agricultural landscape of northern Kerala, India, continues to face severe challenges due to diseases and nutrient deficiencies. Timely identification of these issues is vital for sustaining productivity and minimizing economic losses. This study proposes an advanced deep learning framework comprising ensemble-based classification models for automated detection of black pepper leaf diseases and nutrient deficiencies. Leveraging pre-trained Convolutional Neural Networks (CNNs), six state-of-the-art architectures were fine-tuned and evaluated individually and in ensemble configurations. A custom dataset, the KSD-BPLDND Dataset, containing 8,469 field images across ten classes, was developed to facilitate model training and evaluation. To assess the generalization capability of the proposed system, we additionally validated the models on an external benchmark–the Potato Leaf Disease Dataset comprising 3,076 images across seven disease categories. The cross-dataset evaluation confirms the robustness and transferability of the ensemble models beyond the target crop. Experimental results demonstrate that the majority voting ensemble (BPLDNDNet-MV) achieves 99.53% accuracy on the KSD-BPLDND dataset and maintains high performance on the external potato dataset, reinforcing the general-purpose applicability of the approach. The early fusion ensemble (BPLDNDNet-EF) also achieved strong results with 98.99% accuracy. These findings highlight the potential of the proposed frameworks to support scalable, reliable plant health monitoring in diverse agricultural scenarios.