GCS-YOLO: A Lightweight Detection Algorithm for Grape Leaf Diseases Based on Improved YOLOv8
In view of the issues of high complexity, significant computational resource consumption, and slow inference speed in the detection algorithm for grape leaf diseases, this paper proposes GCS-YOLO, a lightweight detection algorithm based on an improved YOLOv8. The lightweight feature extraction modul...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
MDPI AG
2025-04-01
|
| Series: | Applied Sciences |
| Subjects: | |
| Online Access: | https://www.mdpi.com/2076-3417/15/7/3910 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| Summary: | In view of the issues of high complexity, significant computational resource consumption, and slow inference speed in the detection algorithm for grape leaf diseases, this paper proposes GCS-YOLO, a lightweight detection algorithm based on an improved YOLOv8. The lightweight feature extraction module C2f-GR is proposed to replace the C2f module. C2f-GR achieves lightweight design while effectively capturing detailed features of multi-scale information by replacing partial convolutions in C2f with Ghost Modules. Additionally, RepConv is incorporated into C2f-GR to avoid the complexity of multi-branch structures and enhance gradient flow capability. The CBAM attention mechanism is added to the model to improve the extraction of subtle features of lesions in complex environments. Cross-scale shared convolution parameters and separated batch normalization techniques are used to optimize the detection head, achieving a lightweight design and improving the detection efficiency of the algorithm. Experimental results indicate that the improved model has a number of parameters and computational load of 1.63 M and 4.5 G, respectively, with a mean average precision (mAP@0.5) of 96.2% and a model size of only 3.5 MB. The number of parameters and computational load of the improved model have been reduced by 45.7% and 45.1%, respectively, compared to the baseline model, while the mAP has increased by 1.3%. This lightweight design not only ensures detection accuracy to meet the real-time detection needs of grape leaf diseases but is also more suitable for edge deployment, demonstrating broad application prospects. |
|---|---|
| ISSN: | 2076-3417 |