| 摘 要: 一种提高葡萄叶片病害识别的准确率与实时检测能力的基于改进YOLOv11的葡萄叶片病害检测方法。本文自建葡萄叶片数据集,在原有网络结构基础上进行多层次优化,其中在主干网络末端引入动态蛇形卷积以增强原模型对复杂几何形态病斑的表征能力；在特征提取与融合阶段引入CBAM注意力机制,强化病斑关键区域的特征响应；采用ShuffleNetV2对原模型中的C3K2模块进行轻量化替换,降低参数量和计算复杂度的同时又保持检测性能；最后在检测头中引入ASFF自适应空间特征融合模块,缓解多尺度特征不一致问题。实验基于包含8231张葡萄叶片图像的数据集,其包含黑腐病、黑麻疹病和叶斑病三类主要病害。改进模型在测试集上的mAP@0.5达到95.5%,较原始YOLOv11提升2.90个百分点,推理速度达到75.69FPS,其能够快速,准确检测出这三种葡萄叶片病害,并满足田间复杂环境下的实时检测需求。研究结果表明,该方法在提升葡萄叶片病害检测精度与稳定性的同时兼顾计算效率,为病害智能监测与精准防控提供了技术支撑。 |
| 关键词: 葡萄叶片病害 YOLOv11 深度学习 图像识别 目标检测 智慧农业 |
|
中图分类号:
文献标识码:
|
| 基金项目: 甘肃省高等学校产业支撑项目（2023CYZC-54)；甘肃省林业科学研究院项目（LKY20250644）；甘肃省科技创新人才项目（25RCKAO15 )；兰州市科学技术局：第四批科技计划项目（2025-4-009） |
|
| Grape Leaf Disease Detection Method Based on Improved YOLOv11 |
|
lichenbo, Wei Linjing
|
Gansu Agricultural University
|
| Abstract: An Improved YOLOv11-Based Grape Leaf Disease Detection Method for Enhancing Accuracy and Real-Time Detection Capabilities.Several targeted improvements are incorporated into the original network architecture. Dynamic Snake Convolution is introduced at the end of the backbone to enhance the representation of irregular and fine-grained lesion structures. The CBAM attention mechanism is embedded into the feature extraction and fusion stages to strengthen the model’s focus on disease-relevant regions. In addition, the original C3K2 module is replaced with a lightweight ShuffleNetV2-based structure to reduce computational complexity while maintaining detection performance. Furthermore, an Adaptive Spatial Feature Fusion (ASFF) detection head is adopted to alleviate inconsistencies among multi-scale features.Experiments were conducted on a dataset containing 8,231 grape leaf images covering three major diseases: black rot, black measles, and leaf spot.The results show that the improved YOLOv11 model achieves an mAP@0.5 of 95.5%, representing a 2.90 percentage point improvement over the baseline YOLOv11, with an inference speed of 75.69 FPS. it can rapidly and accurately detect these three grape leaf diseases while meeting real-time detection demands in complex field environments.These results indicate that the proposed method effectively balances detection accuracy and computational efficiency, demonstrating strong potential for practical applications in intelligent disease monitoring and precision agriculture. |
| Keywords: Grape Leaf Diseases YOLOv11 Deep Learning Image Recognition Object Detection Smart Agriculture |
