Scholarlink

<p>Pest insects pose a significant threat to agricultural crops, impacting food security and economic stability, which makes accurate pest classification essential. However, previous methods suffer from limited classification efficiency, overfitting due to large datasets, and reliance on costly agricultural experts. To address these issues, this study introduces the Fruitful Streamlined Attention Network (FSAN) for pest detection and classification. The proposed model was evaluated using the IP102 dataset, which includes 75,000 + images across 102 categories of insect pests affecting eight crop types (rice, corn, wheat, beet, alfalfa, Vitis, citrus, and mango). The dataset was split into 70% for training, 15% for testing, and 15% for validation. To enhance image quality, Luminous Bi-Histogram Equalization (LBHE) was applied in preprocessing, improving contrast and reducing noise. The FSAN architecture integrates seven Streamline Attention-MB (SA-MB) convolution layers with Swish activation, global average pooling, dense layers, and a softmax layer for classification. The FSAN achieved exceptional performance, with 98.88% accuracy, 98.99% precision, 98% recall, and an F1-score of 98.5%. Compared to other models, including DenseNet121, EfficientNet-B0, VGG19, ResNet-50, ResNeSt-50, and ResNeXt-50, FSAN demonstrated superior accuracy and efficiency. FSAN offers a robust and efficient solution for pest detection. This advancement enhances agricultural pest management, aiding in early identification and reducing crop damage.</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

Streamlined attention for insect pest classification: leveraging FSAN

  • 期刊论文

  • 发表日期：

    2025年6月13日

查看全文
  • D. Mansoor Hussain,
  • A. Benazir Begum,
  • N Karthikeyan

摘要

Pest insects pose a significant threat to agricultural crops, impacting food security and economic stability, which makes accurate pest classification essential. However, previous methods suffer from limited classification efficiency, overfitting due to large datasets, and reliance on costly agricultural experts. To address these issues, this study introduces the Fruitful Streamlined Attention Network (FSAN) for pest detection and classification. The proposed model was evaluated using the IP102 dataset, which includes 75,000 + images across 102 categories of insect pests affecting eight crop types (rice, corn, wheat, beet, alfalfa, Vitis, citrus, and mango). The dataset was split into 70% for training, 15% for testing, and 15% for validation. To enhance image quality, Luminous Bi-Histogram Equalization (LBHE) was applied in preprocessing, improving contrast and reducing noise. The FSAN architecture integrates seven Streamline Attention-MB (SA-MB) convolution layers with Swish activation, global average pooling, dense layers, and a softmax layer for classification. The FSAN achieved exceptional performance, with 98.88% accuracy, 98.99% precision, 98% recall, and an F1-score of 98.5%. Compared to other models, including DenseNet121, EfficientNet-B0, VGG19, ResNet-50, ResNeSt-50, and ResNeXt-50, FSAN demonstrated superior accuracy and efficiency. FSAN offers a robust and efficient solution for pest detection. This advancement enhances agricultural pest management, aiding in early identification and reducing crop damage.