Related Experiment Video
Updated: Jan 12, 2026
04:21
Author Spotlight: Sieving Fruit Pulp to Detect Immature Tephritid Fruit Flies in the Field
Published on: July 28, 2023
2.5K
YOLO-lychee-advanced: an optimized detection model for lychee pest damage based on YOLOv11
Xianjun Wu1, Xueping Su1, Zejie Ma1
1Guangdong University of Petrochemical Technology, Maoming, China.
Frontiers in Plant Science
|November 7, 2025
Summary
We developed YOLO-Lychee-advanced, a precise AI detector for identifying lychee stem-borer damage. This advanced model improves pest detection accuracy and speed for fruit quality control.
Area of Science:
- Computer Vision
- Agricultural Technology
- Machine Learning
Background:
- Lychee production faces significant losses due to stem-borer damage.
- Accurate and efficient detection of pest damage is crucial for maintaining fruit quality and marketability.
- Existing detection methods may struggle with fine-grained pest hole features and varying illumination conditions.
Purpose of the Study:
- To develop a lightweight, high-precision object detection model for identifying lychee stem-borer damage on fruit surfaces.
- To enhance the model's ability to detect small pest holes (≤2 mm) and overcome interference from complex fruit peel textures.
- To improve the model's robustness against illumination variations and optimize bounding box regression for accurate damage localization.
Main Methods:
- The study introduces YOLO-Lychee-advanced, built upon the YOLOv11 architecture.
- Key modifications include a C2f module with dual-branch residual connections for fine-grained feature extraction and a CBAM attention block to mitigate peel-texture interference.
- The model utilizes CIoU loss for precise bounding box regression, and data augmentation techniques simulate diverse lighting conditions, expanding the dataset to 9,183 images with a "pest-hole only" annotation strategy.
Main Results:
- YOLO-Lychee-advanced achieved 92.2% precision, 85.4% recall, and 91.7% mAP50 on the test set.
- The model demonstrated superior performance compared to YOLOv9t and YOLOv10n, with mAP50-95 scores of 61.6%, outperforming baselines by 3.4% and 1.7% respectively.
- Real-time detection speed of 37 FPS was maintained, and post-processing optimization further increased precision to 95.5%.
Conclusions:
- YOLO-Lychee-advanced offers a significant advancement in automated detection of lychee stem-borer damage.
- The model's lightweight design, high precision, and real-time performance make it suitable for practical agricultural applications.
- The provided dataset and visualization tool facilitate further research and development in fruit pest detection.
