基于深度学习的多模态病虫害检测模型训练与验证研究  

Training and validation of a multimodal pest detection model based on deep learning

作　　者：李冰[1] 王瀛龙 LI Bing;WANG Yinglong(College of Modern Information Technology,Henan Polytechnic,Zhengzhou 450046,China;Special Equipment Safety Inspection and Research Institute of Henan Province,Zhengzhou 450046,China)

机构地区：[1]河南职业技术学院现代信息技术学院,郑州450046 [2]河南省特种设备检验技术研究院,郑州450046

出　　处：《计算机应用文摘》2024年第22期87-90,96,共5页

基　　金：河南省科技厅科技攻关(242102111190);河南省高等学校重点科研项目(24B520017)。

摘　　要：多模态病虫害数据集具有多样性,导致模型平均精度较低。基于此,文章设计了一种基于深度学习的多模态病虫害检测模型。首先,采集高清图像、音频、视频等多模态数据来整合现有资源构成数据集;其次,进行数据增强以提高模型的泛化能力,减少了过拟合风险;最后,选用深度学习中的MobileNet作为主干网络来捕捉关键特征,模型训练融合了多模态数据,并通过动态权重分配和注意力机制显著增强了检测效能。实验结果表明,在迭代60次时,设计模型的mAP达到0.87,远超文献[1]和文献[2]模型的0.72与0.79,能够快速收敛并准确检测多种病虫害。The diversity of multi modal pest and disease data sets leads to low average accuracy of the model.Based on this paper,a multi modal pest and disease detection model based on deep learning is designed.Firstly,multi modal data such as HD image,audio and video are collected,and existing resources are integrated to form a data set.Secondly,data enhancement is carried out to improve the generalization ability of the model and reduce the risk of overfitting.Finally,MobileNet in deep learning is selected as the backbone network to capture key features,and the model training incorporates multi modal data and significantly enhances detection performance through dynamic weight allocation and attention mechanisms.The experimental results show that after 60 iterations,the mAP of the design model reaches 0.87,which is far higher than the 0.72 and 0.79 of the model in literature[1]and[2].It is proved that the design model can rapidly converge and accurately detect a variety of pests and diseases.

关 键 词：深度学习 MobileNet 多模态 病虫害检测 模型训练 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]