基于卷积神经网络的黄瓜白粉病智能识别算法研究  

Research on intelligent recognition algorithm of cucumber powdery mildew based on convolutional neural network

作　　者：旦真旺姆 全淼儿 钱婷婷[2] 石称华 刘哲辉 常丽英[1] Danzhenwangmu;QUAN Miaoer;QIAN Tingting;SHI Chenghua;LIU Zhehui;CHANG Liying(School of Agriculture and Biology,Shanghai Jiao Tong University,Shanghai 200240,China;Shanghai Academy of Agricultural Sciences,Shanghai 201403,China;Shanghai Kingnew Information Technology Co.,Ltd.,Shanghai 200333,China)

机构地区：[1]上海交通大学农业与生物学院,上海200240 [2]上海市农业科学院,上海201403 [3]上海劲牛信息技术有限公司,上海200333

出　　处：《上海农业学报》2023年第2期127-132,共6页Acta Agriculturae Shanghai

基　　金：上海市农委资助项目(2019-02-08-00-10-F01115)。

摘　　要：采用AlexNet、VGG16、GoogLeNet和ResNet50等4种CNN模型对黄瓜4个病害级别的白粉病叶片图像进行反复迭代训练,探究迭代次数、BATCH_SIZE参数对4种模型识别分类效果的影响,分析不同CNN模型的性能,以选择出应用于黄瓜白粉病识别的最优模型。结果表明:从训练集损失函数的损失率、识别准确率及训练时间综合考量,在当前试验样本条件下,迭代次数为40次,BATCH_SIZE值等于90时,ResNet50模型结果最优,其训练用时为24 min,模型识别准确率为91.30%,对黄瓜白粉病不同病害级别智能识别具有较好的分类性能。Four CNN models,AlexNet,VGG16,GoogLeNet,and ResNet50,were used to repeatedly iterate and train cucumber powdery mildew leaf images of 4 disease levels.The effects of iteration times and BATCH_SIZE parameters on the recognition and classification effects of the 4 models were explored,and the performance of different CNN models was analyzed to select the optimal model for cucumber powdery mildew recognition.The results showed that,considering the loss rate,recognition accuracy,and training time of the loss function in the training set,under the current experimental sample conditions,when the number of iterations was 40 and the BATCH_SIZE was 90,the ResNet50 model had the best results.The training time was 24 minutes,and the accuracy rate of model recognition was 91.30%.It had good classification performance for intelligent recognition of different disease levels of cucumber powdery mildew.

关 键 词：黄瓜白粉病 卷积神经网络 模型 深度学习 智能识别算法 

分 类 号：S126[农业科学—农业基础科学] S436.421