Scholarlink

<p>Image-based disease diagnosis in agricultural fields involves processing heterogeneous visual patterns under strict computational constraints. In coconut plantations, this challenge is further amplified by the need to maintain high predictive performance while reducing computational and memory requirements. This study addresses these challenges by presenting a&#xa0;computationally efficient hybrid feature-based disease detection framework designed for field-level operation.</p><p>The proposed framework integrates multiple visual feature representations, including handcrafted color and texture descriptors and deep feature embeddings, to capture disease-specific characteristics from coconut images. To limit computational overhead while preserving discriminative capability, a&#xa0;quantum-inspired Grey Wolf Optimization strategy is applied for feature selection, leading to a&#xa0;substantial reduction in feature dimensionality.Hyperparameter tuning is further carried out using a&#xa0;multi-objective optimization strategy that jointly considers classification accuracy, macro F1-score, model size, and inference efficiency during model selection.</p><p>The framework is evaluated on a&#xa0;publicly available coconut disease dataset using several machine learning classifiers, including Support Vector Machine, Random Forest, Logistic Regression, and XGBoost. Experimental results show that the Support Vector Machine achieves the best performance, with an accuracy of 98.30%, a&#xa0;classification-stage model size of 802 KB (SVM classifier only), and a&#xa0;classification-stage inference time of 0.017 s measured under GPU conditions. Ablation experiments validate the role of feature fusion and optimization in improving both accuracy and computational efficiency.</p><p>The findings demonstrate that reliable coconut disease detection can be achieved using a&#xa0;lightweight classification component that delivers competitive performance under controlled GPU conditions. The proposed framework highlights the potential of resource-efficient hybrid feature-based image classification for large-scale plantation monitoring and future precision agriculture applications.</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

Hybrid Feature-Based Learning Framework for Computationally Efficient Coconut Disease Diagnosis

  • 期刊论文

  • 发表日期：

    2026年6月25日

查看全文
  • P. S. Pravitha,
  • V. Kumar

摘要

Image-based disease diagnosis in agricultural fields involves processing heterogeneous visual patterns under strict computational constraints. In coconut plantations, this challenge is further amplified by the need to maintain high predictive performance while reducing computational and memory requirements. This study addresses these challenges by presenting a computationally efficient hybrid feature-based disease detection framework designed for field-level operation.

The proposed framework integrates multiple visual feature representations, including handcrafted color and texture descriptors and deep feature embeddings, to capture disease-specific characteristics from coconut images. To limit computational overhead while preserving discriminative capability, a quantum-inspired Grey Wolf Optimization strategy is applied for feature selection, leading to a substantial reduction in feature dimensionality.Hyperparameter tuning is further carried out using a multi-objective optimization strategy that jointly considers classification accuracy, macro F1-score, model size, and inference efficiency during model selection.

The framework is evaluated on a publicly available coconut disease dataset using several machine learning classifiers, including Support Vector Machine, Random Forest, Logistic Regression, and XGBoost. Experimental results show that the Support Vector Machine achieves the best performance, with an accuracy of 98.30%, a classification-stage model size of 802 KB (SVM classifier only), and a classification-stage inference time of 0.017 s measured under GPU conditions. Ablation experiments validate the role of feature fusion and optimization in improving both accuracy and computational efficiency.

The findings demonstrate that reliable coconut disease detection can be achieved using a lightweight classification component that delivers competitive performance under controlled GPU conditions. The proposed framework highlights the potential of resource-efficient hybrid feature-based image classification for large-scale plantation monitoring and future precision agriculture applications.