Detecting and Tracking Maize Leaf and Stalk Growth Using Deep Learning
Author Information
Author(s): Bingwen Liu, Chang Jianye, Hou Dengfeng, Pan Yuchen, Li Dengao, Ruan Jue
Primary Institution: Taiyuan University of Technology
Hypothesis
Can deep learning methods effectively recognize and track maize leaf and stalk trajectories in RGB images?
Conclusion
The study developed a deep learning method called Point-Line Net that achieved an object detection accuracy of 81.5% for recognizing maize leaf and stalk trajectories.
Supporting Evidence
- The Point-Line Net model achieved an object detection accuracy of 81.5%.
- A new lightweight keypoint detection branch was introduced, achieving a magnitude of 33.5.
- The study utilized over 30,000 high-resolution RGB images for training.
- Soft-NMS and D IoU techniques improved detection accuracy in occluded scenarios.
- Keypoint detection was adapted to account for the unique characteristics of maize leaf and stalk trajectories.
Takeaway
This study shows how computers can help farmers by using pictures to count and track the growth of maize leaves and stalks, making farming easier.
Methodology
The study used a deep learning model based on the Mask R-CNN framework to analyze maize images and detect leaf and stalk trajectories.
Potential Biases
Potential bias due to reliance on specific image datasets and annotation methods.
Limitations
The method relies on bounding boxes for detection, which may limit accuracy, and it does not fully address occlusion issues.
Participant Demographics
The study focused on maize plants, with images collected from various growth stages.
Statistical Information
P-Value
0.0001
Confidence Interval
Not applicable
Statistical Significance
p<0.05
Digital Object Identifier (DOI)
Want to read the original?
Access the complete publication on the publisher's website