Scholarlink

<p>Early and precise identification of plant diseases is essential for enhancing agricultural yields and reducing plant loss. Classical diagnosis techniques are usually time-consuming, expensive, and ineffective. This research presents a new technique, the Auto-encoded Attention Artificial Protozoa Convolutional neural network with Few-Shot Learning (3APC-FSL), intended to improve plant disease detection at low training data levels and high precision. The 3APC-FSL model integrates ResNet50 as the backbone for feature extraction and combines it with a ProtoNet Vision Transformer to facilitate robust feature abstraction through attention-based embedding. An Artificial Protozoa Optimizer (APO) is employed to optimize cross-entropy loss, improve training stability, and reduce computational demand, enabling faster and more efficient processing. The interpretability analysis of the 3APC-FSL is analyzed via Gradient-weighted Class Activation Mapping (Grad-CAM) visualizations. The proposed system demonstrates superior performance across multiple datasets, achieving 99.1% accuracy on the Plant Village dataset, 98.9% on New Plant Diseases, and 98.4% on the Plant Doc dataset. These results significantly outperform conventional methods in terms of both accuracy and efficiency. 3APC-FSL model proves effective across various datasets, achieving 99.1% accuracy on the Plant Village dataset, 98.9% on the New Plant Diseases dataset, and 98.4% on the Plant Doc dataset. These outcomes significantly surpass traditional methods in both accuracy and efficiency.</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

Auto-Encoded Attention Artificial Protozoa Convolutional Neural Networks with Few-Shot Learning for Plant Disease Detection

  • 期刊论文

  • 发表日期：

    2025年11月1日

查看全文
  • Binal Upadhyay,
  • Amit Ganatra

摘要

Early and precise identification of plant diseases is essential for enhancing agricultural yields and reducing plant loss. Classical diagnosis techniques are usually time-consuming, expensive, and ineffective. This research presents a new technique, the Auto-encoded Attention Artificial Protozoa Convolutional neural network with Few-Shot Learning (3APC-FSL), intended to improve plant disease detection at low training data levels and high precision. The 3APC-FSL model integrates ResNet50 as the backbone for feature extraction and combines it with a ProtoNet Vision Transformer to facilitate robust feature abstraction through attention-based embedding. An Artificial Protozoa Optimizer (APO) is employed to optimize cross-entropy loss, improve training stability, and reduce computational demand, enabling faster and more efficient processing. The interpretability analysis of the 3APC-FSL is analyzed via Gradient-weighted Class Activation Mapping (Grad-CAM) visualizations. The proposed system demonstrates superior performance across multiple datasets, achieving 99.1% accuracy on the Plant Village dataset, 98.9% on New Plant Diseases, and 98.4% on the Plant Doc dataset. These results significantly outperform conventional methods in terms of both accuracy and efficiency. 3APC-FSL model proves effective across various datasets, achieving 99.1% accuracy on the Plant Village dataset, 98.9% on the New Plant Diseases dataset, and 98.4% on the Plant Doc dataset. These outcomes significantly surpass traditional methods in both accuracy and efficiency.