基于改进EfficientNet-Lite适用于复杂背景下茶树病害识别轻量级移动模型  

A Lightweight Mobile Model Based on Improved EfficientNet-Lite for Tea Plant Disease Recognition in Complex Backgrounds

作　　者：梁俊杰 王东霞 Liang Junjie;Wang Dongxia(College of Mathematics and Computer Science,Zhejiang A&F University,Hangzhou 311300,Zhejiang,China;Shanxi Normal University,Taiyuan 041081,Shanxi,China)

机构地区：[1]浙江农林大学数学与计算机科学学院,浙江杭州311300 [2]山西师范大学,山西太原041081

出　　处：《农业技术与装备》2024年第9期23-25,28,共4页Agricultural Technology & Equipment

摘　　要：提出了一种轻量化的茶树病害实时检测模型EGNet,使用EfficientNet-Lite0作为骨干网络,引入ECA注意力机制和Ghost模块,并通过AdaBelief优化算法对参数进行优化。结果表明,在自建茶树病害数据集上的识别准确率达到97.25%,优于其他基线模型。同时在Mini-ImageNet和IP102数据集上也表现优异,证明模型具有良好的鲁棒性。在此基础上,进一步开发了一套茶树病害识别平台与手机App,可在智能手机上分别进行离线检测和在线检测,具有实际应用价值。This paper proposed a lightweight real-time detection model for tea diseases,EGNet,using EfficientNet-Lite0 as the backbone network,introducing ECA attention mechanism and Ghost module,and optimizing parameters through AdaBelief optimization algorithm.The experimental results showed that the recognition accuracy on the self built tea disease dataset reaches 97.25%,which was better than other baseline models.At the same time,it also performed well on the Mini ImageNet and IP102 datasets,proving that the model has good robustness.On this basis,a tea disease recognition platform and mobile App have been further developed,which could perform offline and online detection on smartphones respectively,and have practical application value.

关 键 词：EGNet 茶树病害 实时检测 轻量化模型 注意力机制 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]