Related Experiment Video
Updated: Jun 21, 2025
11:49
Cereal Crop Ear Counting in Field Conditions Using Zenithal RGB Images
Published on: February 2, 2019
9.3K
Weed detection and recognition in complex wheat fields based on an improved YOLOv7.
Kaixin Wang1, Xihong Hu1, Huiwen Zheng1
1College of Electronic Engineering, South China Agricultural University, Guangzhou, China.
Frontiers in Plant Science
|July 9, 2024
Summary
A new CSCW-YOLOv7 model accurately detects five types of weeds in wheat fields, even when they are occluded or small. This improved deep learning approach enhances weed management precision and efficiency in agriculture.
Area of Science:
- Agricultural Science
- Computer Vision
- Machine Learning
Background:
- Precise weed detection is crucial for effective weed management in agriculture.
- Similar visual characteristics (color, morphology) and occlusion between crops and weeds present significant detection challenges.
- Existing methods struggle with accurate identification in complex field environments.
Purpose of the Study:
- To propose an improved YOLOv7 architecture, CSCW-YOLOv7, for accurate identification and classification of five common weed types in wheat fields.
- To enhance the detection of small and occluded weeds in complex agricultural settings.
Main Methods:
- A dataset of five common weed species was created.
- The CSCW-YOLOv7 model incorporated CARAFE for improved small target recognition, Squeeze-and-Excitation (SE) networks for feature enhancement, contextual transformer (CoT) modules for global information capture, and a Wise Intersection over Union (WIoU) loss function for better bounding box prediction of occluded weeds.
- Ablation experiments and comparisons with baseline YOLOv7 and other deep learning models were conducted.
Main Results:
- CSCW-YOLOv7 achieved high performance metrics: 97.7% accuracy, 98% recall, and 94.4% mean average precision (mAP).
- Compared to baseline YOLOv7, CSCW-YOLOv7 demonstrated improvements in precision (+1.8%), recall (+1%), and mAP (+2.1%), with a 10.7% reduction in parameters and 10% fewer FLOPs.
- Visualizations confirmed the model's ability to learn representative features for detecting weeds of various scales, outperforming other models in distinguishing overlapped and small-scale weeds.
Conclusions:
- The proposed CSCW-YOLOv7 model significantly enhances weed detection accuracy and efficiency in complex wheat fields.
- The model's ability to handle small and occluded weeds makes it a promising tool for practical agricultural applications.
- CSCW-YOLOv7 offers a computationally efficient solution with potential for real-world deployment in precision agriculture.
