Scholarlink

The early and accurate diagnosis of potato leaf diseases is vital for mitigating agricultural losses and ensuring food security. This study proposes a lightweight and efficient hybrid framework that integrates deep feature extraction using a pre-trained DenseNet121 model, dimensionality reduction via Principal Component Analysis (PCA), and Random Forest (RF) algorithm classification. With high-level feature extraction from images of potato leaves, feature space compression via PCA, and employing a robust RF classifier, the system achieves high accuracy with computational efficiency. The experiment results show that the DenseNet121 + PCA + RF model achieves a test accuracy of 98.65%, which is significantly higher than the baseline Random Forest model using handcrafted features. Additionally, the proposed model demonstrates superior discriminative capability, with macro- and micro-average Area Under the Curve (AUC) values of 0.994 and 0.995, respectively. Comparative analyses highlight the advantages of the hybrid approach in terms of interpretability, reduced feature dimensionality, and simplicity of deployment in low-resource agricultural environments. This study presents a high-performance and scalable solution for precision agriculture, providing actionable insights into the development of AI-driven plant disease diagnostic systems for real-world applications.

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

A Lightweight Deep Feature Fusion and PCA-Optimized Random Forest Framework for Multi-class Potato Leaf Disease Classification

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Roseline Oluwaseun Ogundokun,
  • Pius Adewale Owolawi,
  • Etienne A. van Wyk

摘要

The early and accurate diagnosis of potato leaf diseases is vital for mitigating agricultural losses and ensuring food security. This study proposes a lightweight and efficient hybrid framework that integrates deep feature extraction using a pre-trained DenseNet121 model, dimensionality reduction via Principal Component Analysis (PCA), and Random Forest (RF) algorithm classification. With high-level feature extraction from images of potato leaves, feature space compression via PCA, and employing a robust RF classifier, the system achieves high accuracy with computational efficiency. The experiment results show that the DenseNet121 + PCA + RF model achieves a test accuracy of 98.65%, which is significantly higher than the baseline Random Forest model using handcrafted features. Additionally, the proposed model demonstrates superior discriminative capability, with macro- and micro-average Area Under the Curve (AUC) values of 0.994 and 0.995, respectively. Comparative analyses highlight the advantages of the hybrid approach in terms of interpretability, reduced feature dimensionality, and simplicity of deployment in low-resource agricultural environments. This study presents a high-performance and scalable solution for precision agriculture, providing actionable insights into the development of AI-driven plant disease diagnostic systems for real-world applications.