Related Experiment Video
Updated: Jul 2, 2025
08:47
Author Spotlight: UAV Remote Sensing for Efficient Invasive Plant Biomass Estimation
Published on: February 9, 2024
1.4K
Automatic detection method for tobacco beetles combining multi-scale global residual feature pyramid network and
Yuling Chen1,2, Xiaoxia Li1,3, Nianzu Lv4
1School of Information Engineering, Southwest University of Science and Technology, Mianyang, 621010, Sichuan, China.
Scientific Reports
|February 28, 2024
Summary
This study introduces an advanced method for detecting tobacco beetles in images, even with low pixel counts and noise. The new approach significantly improves detection accuracy and recall rates for pest identification.
Area of Science:
- Agricultural Entomology
- Computer Vision
- Machine Learning
Background:
- Identifying storage pests like tobacco beetles from images is challenging due to low object pixel counts and high image noise.
- Existing methods often struggle with information scarcity and lack of identifiable features in such conditions.
Purpose of the Study:
- To propose an automatic monitoring method for tobacco beetles using deep learning.
- To enhance the accuracy and reliability of tobacco beetle detection in challenging image conditions.
Main Methods:
- Developed a Multi-scale Global residual Feature Pyramid Network (MGrFPN) to capture rich semantic and low-level features, reducing missed detections.
- Designed a Dual-path Deformable receptive field Guided Attention Module (DDrGAM) for improved feature fusion and localization accuracy by considering spatial-geometric deformations.
- Created a new multi-scene tobacco beetle dataset with 28,080 labeled images to simulate real-world scenarios.
Main Results:
- The proposed MGrFPN-DDrGAM method, within the Faster R-CNN framework, achieved a detection precision of 91.4% and recall of 98.4% at an IoU of 0.5.
- At an IoU of 0.7, detection precision improved by 32.9% compared to Faster R-CNN and 6.9% compared to FPN.
- The method demonstrated superior performance over current mainstream detection techniques.
Conclusions:
- The MGrFPN-DDrGAM method effectively addresses the challenges of detecting small and noisy tobacco beetle images.
- The proposed approach offers a significant advancement in automated pest monitoring systems.
- This work provides a robust solution for agricultural pest identification using computer vision.
