基于无锚框检测网络的茶叶嫩芽识别方法研究  

Research based on recognition method for tea buds based on anchor free detection network

作　　者：毛腾跃[1] 朱俊杰 帖军[1] MAO Tengyue;ZHU Junjie;TIE Jun(College of Computer Science&Hubei Provincial Engineering Research Center for Intelligent Management of Manufacturing Enterprises,South-Central Minzu University,Wuhan 430074,China)

机构地区：[1]中南民族大学计算机科学学院&湖北省制造企业智能管理工程技术研究中心,武汉430074

出　　处：《中南民族大学学报（自然科学版）》2023年第4期489-496,共8页Journal of South-Central University for Nationalities：Natural Science Edition

基　　金：湖北省技术创新专项重大项目(2019ABA101);湖北省科技计划项目(2019CFC890)。

摘　　要：利用机器学习方法进行茶叶嫩芽识别有助于茶叶生产的全程智能化.目前的嫩芽识别方法依赖复杂的预处理,导致嫩芽识别效率普遍不高.基于无锚框检测网络CenterNet,提出了一种无需预处理的高速茶叶嫩芽识别方法.针对CenterNet特征提取网络规模过大、识别速度过低的问题,将其特征提取网络替换为ResNet-50.利用改进后的CenterNet模型识别茶叶的一芽、一芽一叶和一芽二叶部分,得到模型的精确度为83.1%,召回率为85.4%,mAP为87.7%,识别效果优于同类无预处理识别方法.It is helpful to the whole intellectualization of tea production that using machine learning methods to recognize tea buds.At present,the methods of tea buds recognition based on complex preprocessing,which results in low recognition efficiency for tea buds.Based on CenterNet,a method that no pretreatment needed to recognize tea bud is proposed.For CenterNet’s oversize scale and low speed,the proposed method replaced CenterNet’s feature extraction network with ResNet-50.Recognizing the parts one bud,one bud,one leaf and one bud,two leaf of tea by improved CenterNet.The precision of tea bud recognition model is 83.1%,recall rate is 85.4%,and mAP is 87.7%.The recognize effect is better than similar methods without preprocessing.

关 键 词：目标检测 深度学习 无锚框检测网络 茶叶嫩芽 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]