IMpc-PyrYOLO: Hybrid YOLO Based Feature Pyramidal Network for Pest Detection in Rice Leaves

Faster R-CNN Faster-PestNet Scalable and Efficient Network Gaussian Filtering Channel Attention Mechanism and Batch Normalization

Authors

  • Ramesh Shunmugam 1) Center of Intelligent Cloud Computing (CICC), CoE of Advanced Cloud, Multimedia University, Melaka, Malaysia. 2) Department of Mechatronics Engineering, Rajalakshmi Engineering College, Chennai, 602105, India
  • Sumendra Yogarayan
    sumendra@mmu.edu.my
    1) Center of Intelligent Cloud Computing (CICC), CoE of Advanced Cloud, Multimedia University, Melaka, Malaysia. 3) Faculty of Information Science and Technology, Multimedia University, Jln Ayer Keroh Lama, Melaka, 75450, Malaysia
  • Siti Fatimah Abdul Razak 1) Center of Intelligent Cloud Computing (CICC), CoE of Advanced Cloud, Multimedia University, Melaka, Malaysia. 3) Faculty of Information Science and Technology, Multimedia University, Jln Ayer Keroh Lama, Melaka, 75450, Malaysia
  • Md. Shohel Sayeed 1) Center of Intelligent Cloud Computing (CICC), CoE of Advanced Cloud, Multimedia University, Melaka, Malaysia. 3) Faculty of Information Science and Technology, Multimedia University, Jln Ayer Keroh Lama, Melaka, 75450, Malaysia
Vol. 9 No. 3 (2025): June
Research Articles

Downloads

Pests pose a significant threat to global food security, making early detection crucial for maintaining crop health. Traditional pest detection models suffered from inefficiencies such as long processing time and low accuracy, which hinder effective disease management. In order to overcome these existing issues, a novel improved efficient channel attention mechanism assisted feature pyramidal network-based YOLO model (IMpc-PyrYOLO) for rice leaf pest detection. The model integrates an efficient channel attention (ECA) mechanism with the feature pyramidal network (FPN) in the YOLO network to improve multi-scale feature extraction and pest classification accuracy. Additionally, an upgraded weighted Gaussian wiener filter (Up-weGaf) is employed for noise reduction, improving image clarity. The model is evaluated on two benchmark datasets such as IP_RicePests and IP102. Experimental results demonstrate that IP_RicePests attains a precision of 95.8%, recall of 96% and F1-score of 95.9%, whereas IP102 attains a higher precision of 97.8%, recall of 96%, mean Average Precision (mAP) of 95.9% and Intersection over Union (IoU) of 97% with processing time of 2.5 seconds. The proposed model significantly outperforms existing methods in accuracy and computational efficiency, which provides a robust and scalable solution for real-time pest detection in agriculture.