基于改进MobileNet V3模型的蜜源植物识别与分类  

Nectar plant recognition and classification based on the improved MobileNet V3 model

作　　者：董梦婷 曹浩 赵天 赵旭 尚林薇 DONG Mengting;CAO Hao;ZHAO Tian;ZHAO Xu;SHANG Linwei(College of Mechanical Engineering,Anhui Science and Technology University,Fengyang 233100,China;College of Information and Network Engineering,Anhui Science and Technology University,Bengbu 233000,China)

机构地区：[1]安徽科技学院机械工程学院,安徽凤阳233100 [2]安徽科技学院信息与网络工程学院,安徽蚌埠233000

出　　处：《安徽科技学院学报》2025年第2期75-81,共7页Journal of Anhui Science and Technology University

基　　金：安徽省协同创新项目(GXXT-2023-105)。

摘　　要：针对蜜源植物种类繁多及其复杂的外观特征,提出一种高效且准确的识别和分类方法,旨在提升蜜蜂授粉活动的生态效益以及蜂蜜生产的质量和效率。本文提出一种使用改进版轻量化模型MobileNet V3进行蜜源植物识别和分类的方法。通过数据融合的方式扩展数据集,结合ECA机制优化通道间的特征权重分配,同时引入Mish激活函数以改善模型的非线性特性,提升蜜源植物的识别准确性。改进后的模型识别准确率由原模型的89.59%提升至94.20%,精确度从89.9%提升至94.5%,改进后的模型体积由原来的16 MB优化至10 MB,展现了更为出色的性能与效率。本文提出的改进方法显著提高了蜜源植物识别的准确性和效率并减少了存储和计算资源的占用,对精准农业和蜂蜜生产具有重要的实用价值。In response to the diverse species and complex appearance of nectar source plants,a highly efficient and accurate method for identification and classification is proposed,aiming to enhance the ecological benefits of bee pollination activities and improve the quality and efficiency of honey production.A method is proposed for the identification and classification of nectar source plants using an improved version of the lightweight MobileNet V3 model.The dataset is expanded through data fusion,and the distribution of feature weights between channels is optimized using the ECA mechanism.Additionally,the Mish activation function is introduced to improve the model's non-linear characteristics,enhancing the accuracy of nectar source plant identification.The improved model has shown enhanced performance and efficiency,with the recognition accuracy increased from 89.59%to 94.20%,and precision improved from 89.9%to 94.5%.Additionally,the size of the model has been optimized from 16 MB to 10 MB.The proposed improved method significantly improves the accuracy and efficiency of nectar plant identification and reduces the occupation of storage and computing resources,which has important practical application value for precision agriculture and honey production.

关 键 词：卷积神经网络 注意力机制 精准农业 蜜源识别分类 数据集融合 

分 类 号：S897[农业科学—特种经济动物饲养]