Article Details

Abstract

To address the problems of large variations in target scale, complex backgrounds, and class imbalance in grape disease and pest detection, this study proposes an improved YOLOv10 recognition model. The model introduces a CFM multi-scale feature fusion module and an ECA lightweight attention mechanism into the backbone network, adopts Mobile Conv to achieve a lightweight network, and designs a hybrid loss function combining Hybrid IoU and Focal Loss to enhance the representation capability for small targets and the detection performance on minority-class samples. Experimental results show that compared with the original YOLOv10 model, the improved model increases recognition precision and recall by 6.4 and 8.3 percentage points respectively, with the number of parameters reduced to 20.5 M and the inference speed reaching 78 FPS. Ablation and comparative experiments verify the positive gain of each improved module. Field tests show that the model maintains a detection precision above 87.5% under different lighting conditions and exhibits good robustness and generalization ability, thus providing technical support for intelligent monitoring of grape diseases and pests and precise prevention and control in smart orchards.

Keywords

grape diseases and pests; object detection; YOLOv10; lightweight; multi-scale feature fusion

References

1. [1] Ma Cong, Chen Xuedong, Zhang Xuejian, etc Algorithm for detecting leaf diseases and pests of wine grapes based on improved YOLOv10s [J]. Jiangsu Journal of Agriculture, 2025, 41 (12): 2387-2402
2. [2] Qiao Shicheng, Zhao Chenyu, Li Chengyong, etc A lightweight and complex background grape leaf disease detection method based on improved YOLOv10n [J]. Journal of Shenyang Agricultural University, 2025, 56 (6): 45-54
3. [3] Zhen Yuanyuan, Liu Ping, Jin Yuchun, etc Detection and identification of grape leaf diseases based on improved YOLOv8 [J]. Journal of Agricultural Engineering, 2025, 41 (14): 148-154
4. [4] Zhang Linqian, Bayintana, Zeng Qingsong Early detection method for grape leaf diseases based on StyleGAN2-ADA and improved YOLOv7 [J]. Journal of Agricultural Machinery, 2024, 55 (1): 241-252
5. [5] Cai Yinan, Xiao Xiaoling Lightweight method for grape leaf disease and pest detection model based on improved YOLO v5n [J]. Jiangsu Agricultural Science, 2024, 52 (7): 198-205
6. [6] Lin Wenqing, Li Shuqin Grape leaf disease detection method based on improved RT-DETR [J]. Journal of Agricultural Engineering, 2025, 41 (15): 145-154
7. [7] Shi Chengqi, Hu Yaqi, Zhang Bingtao VitiDetect Net: High precision detection algorithm for grape leaf diseases and pests [J]. Computer Science and Exploration, 2026, 20 (8): 2384-2395
8. [8] Zhai Changyuan, Liu Bohao, Li Cuiling, etc Research progress on detection and grading methods for major grape diseases [J]. Journal of Agricultural Machinery, 2025, 56 (8): 341-359
9. [9] Yang Fan Research on Identification and Detection Technology of Grape Leaf Diseases and Pests Based on Deep Learning [D]. Wuhan: Wuhan University of Light Industry, 2025

Funding

Key Fields Special Project for Ordinary Universities in Guangdong Province in 2024 (2024ZDZX4133); Characteristic Innovation Project for Ordinary Universities in Guangdong Province in 2025 (2025KTSCX320); Special Fund for Guangdong Province's Science and Technology Innovation Strategy in 2026 (pdjh2026bc369)

About the authors

Zheng Shaoxiong (1990-), male, Han ethnicity, from Chaozhou, Guangdong, associate professor, doctoral student, research interests include deep learning.