Related Experiment Video
Updated: May 2, 2026
11:49
Cereal Crop Ear Counting in Field Conditions Using Zenithal RGB Images
Published on: February 2, 2019
9.8K
Exploiting adversarial style for generalized and robust weed segmentation in rice paddy field
Yaoxuan Zhang1, Hao Cai1, Jiahui Ye1
1College of Engineering, South China Agricultural University, Guangzhou, China.
Frontiers in Plant Science
|December 17, 2025
Summary
This study presents a novel deep learning method for weed identification in rice fields. The Style-guided Weed Instance Segmentation (SWIS) method enhances accuracy in precision agriculture.
Area of Science:
- Agricultural Science
- Computer Vision
- Deep Learning
Background:
- Effective weed management is crucial for rice yield and quality in precision agriculture.
- Accurate differentiation between weeds and rice crops is a persistent challenge for precision weeding technologies.
Purpose of the Study:
- To introduce an innovative deep learning methodology for weed identification and segmentation in paddy fields.
- To enhance cross-environment generalization and feature robustness for weed recognition models.
Main Methods:
- Developed a Style-guided Weed Instance Segmentation (SWIS) method.
- Integrated Random Adaptive Instance Normalization (RAIN) for stochastic style transformation.
- Employed Dynamic Gradient Back-propagation (DGB) for adversarial feature optimization.
Main Results:
- Achieved a Weed Intersection over Union (Weed IoU) of 70.49% on field data.
- Demonstrated significant performance improvement over existing comparison methods.
- Validated the method's effectiveness for real-world agricultural applications.
Conclusions:
- The SWIS method offers an effective solution for weed identification and segmentation in precision agriculture.
- This research advances computer vision applications in agriculture.
- Provides a foundation for developing more advanced weed recognition models.
