您好，欢迎访问浙江省农业科学院 机构知识库！
浙江省农业科学院机构知识库

全部

  • 全部
  • 标题
  • 学者
  • 机构
  • 关键词

A Study on Real-Time Detection of Rice Diseases in Farmlands Based on Multidimensional Data Fusion

文献类型： 外文期刊

作者： Ye, Wei 1 ; Jiang, Fei 2 ; Li, Zhaoxing 2 ; Zhao, Lei 2 ; Wang, Jiaoyu 4 ; Wang, Hongkai 5 ;

作者机构： 1.Zhejiang Univ, Coll Control Sci & Engn, Hangzhou 310027, Peoples R China

2.Zhejiang Univ, Huzhou Inst, Huzhou, Peoples R China

3.Huzhou Univ, Sch Engn, Huzhou 313000, Peoples R China

4.Zhejiang Acad Agr Sci, Hangzhou 310021, Peoples R China

5.Zhejiang Univ, Inst Biotechnol, Hangzhou 310058, Peoples R China

关键词： attention mechanism; deep learning; multidimensional feature fusion; object detection; YOLOv5

期刊名称：PLANT DISEASE （ 影响因子：4.4； 五年影响因子：4.8 ）

ISSN： 0191-2917

年卷期： 2025 年 109 卷 6 期

页码：

收录情况： SCI

摘要： To meet the need of crop leaf disease detection in complex scenarios, this study designs a method based on the computing power of mobile devices that ensures both detection accuracy and real-time efficiency, offering significant practical application value. Based on a comparison with existing mainstream detection models, this paper proposes a target detection and recognition algorithm, TG_YOLOv5, which utilizes multidimensional data fusion on the YOLOv5 model. The Triplet Attention mechanism and C3CBAM module are incorporated into the network structure to capture connections between spatial and channel dimensions of input feature maps, thereby enhancing the model's feature extraction capabilities without significantly increasing the parameter count. The GhostConv lightweight module is used to construct the backbone network, reducing model complexity, shrinking the model size, and improving detection speed. A self-constructed rice leaf disease dataset is used for experimentation. Results show that TG_YOLOv5 achieves a mean Average Precision (mAP) of 98.3% and a recall rate of 97.2%, representing a 1.2% improvement in mAP and a 4.3% improvement in recall over the traditional YOLOv5 algorithm. The trained lightweight model is then deployed on a Raspberry Pi using the mobile neural network (MNN) engine for acceleration, showing a 73.8% increase in detection speed across models after MNN acceleration. Additionally, this model achieves satisfactory detection accuracy and speed on apple and tomato datasets, validating its generalization ability. This research provides a theoretical foundation for remote real-time detection of rice diseases in agriculture.

  • 相关文献
作者其他论文 更多>>
置顶
购物车
反馈

© 京ICP备09089781号-29 主办单位：浙江省农业科学院

学术资源： 中国农业科学院 农业专业知识服务系统 农科联盟资源共建共享平台 中国农业科技文献与信息服务平台 中国农业期刊集成服务平台