Related Experiment Video
Updated: Oct 6, 2025
08:47
Author Spotlight: UAV Remote Sensing for Efficient Invasive Plant Biomass Estimation
Published on: February 9, 2024
1.7K
A novel deep learning-based method for detection of weeds in vegetables
Xiaojun Jin1, Yanxia Sun2, Jun Che1
1College of Mechanical and Electronic Engineering, Nanjing Forestry University, Nanjing, China.
Pest Management Science
|January 21, 2022
Summary
This study introduces a deep learning method for precise weed detection in vegetable fields, classifying all green objects not recognized as crops as weeds. YOLO-v3 demonstrated superior accuracy and efficiency for weed identification, paving the way for robotic weed control.
Area of Science:
- Agricultural technology
- Computer vision
- Machine learning
Background:
- Precision weed control is crucial for reducing agricultural inputs in vegetable farming.
- Accurate weed detection is challenging due to diverse weed species, growth stages, and densities.
- Current methods struggle with the complexity of identifying numerous weed types in vegetable fields.
Purpose of the Study:
- To develop a novel deep learning-based method for efficient and accurate weed detection in vegetable fields.
- To simplify weed detection by recognizing crops and classifying all other green objects as weeds.
- To evaluate the performance of different deep learning models for this task.
Main Methods:
- A deep learning approach was employed for weed detection.
- The method identifies vegetable crops and categorizes remaining green objects as weeds.
- Performance was evaluated using YOLO-v3, CenterNet, and Faster R-CNN models.
Main Results:
- Deep learning models achieved an average precision (AP) above 97%.
- YOLO-v3 exhibited the highest accuracy (0.971 F1 score) and computational efficiency.
- Optimal confidence thresholds were determined for each model (YOLO-v3: 0.4, CenterNet: 0.6, Faster R-CNN: 0.4/0.5).
Conclusions:
- Deep learning methods reliably detect weeds in vegetable crops, simplifying the overall detection process.
- The proposed approach effectively reduces weed detection complexity by not requiring specific weed species identification.
- These findings support the advancement of site-specific robotic weed control systems in agriculture.
