Related Experiment Video
Updated: Jan 14, 2026
06:41
Imaging and Analysis for Quantifying Maize (Zea mays) Abiotic Stress Phenotypes
Published on: March 28, 2025
1.5K
Application of real-time detection transformer based on convolutional block attention module and grouped convolution
Yunlong Wu1,2,3, Shouqi Yuan1,3, Yue Tang1
1Research Center of Fluid Machinery Engineering and Technology, Jiangsu University, Zhenjiang, China.
Frontiers in Plant Science
|October 27, 2025
Summary
Accurate maize seedling detection is vital for smart farming. The new CBAM-RTDETR model significantly improves seedling identification in complex field conditions using enhanced feature extraction.
Area of Science:
- Agricultural Engineering
- Computer Vision
- Remote Sensing
Background:
- Intelligent detection and counting of maize seedlings are essential for modern smart agriculture and breeding.
- Challenges in field detection include small target size and complex farmland environments.
Purpose of the Study:
- To develop an improved maize seedling detection model for enhanced accuracy and real-time performance.
- To address the limitations of existing methods in complex agricultural settings.
Main Methods:
- Proposed an improved detection model named CBAM-RTDETR.
- Integrated the CBAM module and grouped convolution into the RT-DETR backbone network for feature extraction.
Main Results:
- Achieved a mean Average Precision (mAP0.5) of 92.9% and a mean Average Recall (AR) of 64.4%.
- Reached a Frames Per Second (FPS) of 87f/s, outperforming comparison models.
- Enhanced shallow edge detail and feature diversity for improved seedling recognition.
Conclusions:
- The CBAM-RTDETR model effectively overcomes challenges in real-time and accurate maize seedling identification.
- The model shows significant potential for application in UAV remote sensing for smart maize cultivation.
