基于百度AI卷积神经网络的茶树结霜智能识别  

Intelligent recognition of tea frost based on Baidu AI convolutional neural network

作　　者：朱兰娟[1] 胡德云[1] 华行祥 李时睿 杨欣洁 张旭晖 ZHU Lanjuan;HU Deyun;HUA Xingxiang;LI Shirui;YANG Xinjie;ZHANG Xuhui(Hangzhou Meteorological Bureau,Hangzhou 310051,China;Faculty of Geographical Science,Beijing Normal University,Beijing 100875,China;Jiangsu Climate Center,Nanjing 210041,China)

机构地区：[1]杭州市气象局,杭州310051 [2]北京师范大学地理科学学部,北京100875 [3]江苏省气候中心,南京210041

出　　处：《气象科学》2023年第2期245-253,共9页Journal of the Meteorological Sciences

基　　金：浙江省重点研发计划项目(2021C02036);浙江省气象科技重点项目(2019ZD16,2019ZD17);国家重点研发计划项目(2019YFD1002203)。

摘　　要：基于采集的近20000张茶园图像,分别筛选出1000张茶园结霜和1000张未结霜图像作为训练样本,利用百度AI开放平台的EasyDL经典版图像识别功能,应用卷积神经网络算法,建立茶园结霜识别模型,进行茶树结霜图像智能判别。茶树结霜智能识别模型的准确率、F1评分、精确率、召回率分别为99.2%、99.2%、99.3%、99.1%。模型检验结果为:样本地20张结霜图像的平均识别置信度为95.51%,20张非结霜图像的平均识别置信度为99.99%;非样本地(江南茶区)80张结霜图像的平均识别置信度为97.53%,80张非结霜图像的平均识别置信度为95.11%。There are 1000 frosted and 1000 non-frosted images of tea gardens were selected as training samples from nearly 20000 images of tea gardens,and the tea garden frost model was established by using the EasyDL image recognition function of Baidu AI open platform in order to intelligently recognize the tea garden frosting.The accuracy rate,F1-score,precision rate and recall rate of the tea garden frost model were 99.2%,99.2%,99.3%and 99.1%,respectively.The model test results showed that the average recognition confidence of 20 frosted images was 95.51%,and the average recognition confidence of 20 non-frosted images was 99.99%.The average recognition confidence of 80 frosted images in the non-sample area(tea area in Jiangnan)was 97.53%,and the average recognition confidence of 80 non-frosted images was 95.11%.

关 键 词：百度AI 卷积神经网络 茶树 结霜图像 智能识别 

分 类 号：P49[天文地球—大气科学及气象学]