PdYOLO: A Lightweight Algorithm for Detecting Peach Fruits Against a Peach Tree Background
Accurate peach fruit detection in a fruit tree environment is of great significance for intelligent harvesting technology and accurate yield prediction. Peaches have a short picking cycle and a concentrated ripening period. However, the backgrounds of peach trees, including trunks, branches, and lea...
Saved in:
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
IEEE
2024-01-01
|
| Series: | IEEE Access |
| Subjects: | |
| Online Access: | https://ieeexplore.ieee.org/document/10786972/ |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | Accurate peach fruit detection in a fruit tree environment is of great significance for intelligent harvesting technology and accurate yield prediction. Peaches have a short picking cycle and a concentrated ripening period. However, the backgrounds of peach trees, including trunks, branches, and leaves, contribute to uneven sample distribution. Additionally, the diverse shapes of peach fruits, variations in light, leaf shading, and dense clustering complicate detection. To address these challenges, this study introduces PdYOLO (Peach detection based on YOLOv8s), an innovative peach detection model optimized within the YOLOv8s framework. This model enhances recognition accuracy, reduces algorithm complexity, and satisfies lightweight requirements. First, the CIOU regression loss function in YOLOv8s is replaced with the WIoUv2 regression loss function, effectively alleviating the negative impact of uneven distribution of positive and negative samples during model training through a more balanced gradient gain distribution strategy, which significantly improves detection accuracy. Second, the CSPPC module replaces the original C2f module in the neck network, reducing the number of parameters and stabilizing model accuracy while lowering computational complexity and resource consumption. Finally, a new head layer P2-SLD was designed, which replaced the original detection head with a shared lightweight detection head SLD, introduced the P2 layer and removed the P5 layer, which improved the lightweight level of the model, improved the processing efficiency and reduced the energy consumption while ensuring the detection accuracy. Experimental results indicate that, compared to the benchmark model YOLOv8s, PdYOLO achieves significant improvements across several key performance indicators: Precision increased by 3.7%, Recall by 10.8%, and mAP@50 by 4.7%. Furthermore, the number of parameters in the model was reduced by 43.3%, from 11.13 million to 6.3 million, fully meeting the lightweight requirements for edge device deployment and providing essential support for intelligent orchard management. |
|---|---|
| ISSN: | 2169-3536 |