基于CNN与LSTM的烟叶烘烤阶段识别方法  

Identification of tobacco flue-curing stages based on CNN and LSTM

作　　者：闫九福 冉茂 朱荣光[2] 王松峰[1] 徐小洪[3] 杨超 张学杰[5] 孙福山[1] 孟令峰 YAN Jiufu;RAN Mao;ZHU Rongguang;WANG Songfeng;XU Xiaohong;YANG Chao;ZHANG Xuejie;SUN Fushan;MENG Lingfeng(Tobacco Research Institute of CAAS,Key Laboratory of Tobacco Biology and Processing,Ministry of Agriculture and Rural Affairs,Qingdao 266101,Shandong,China;College of Mechanical and Electrical Engineering,Shihezi University,Shihezi 832000,Xinjiang,China;Chongqing Tobacco Company of CNTC,Chongqing 400023,China;Pengshui Branch,Chongqing Tobacco Company of CNTC,Pengshui 409600,Chongqing,China;Qianjiang Branch,Chongqing Tobacco Company of CNTC,Qianjiang 409000,Chongqing,China)

机构地区：[1]中国农业科学院烟草研究所农业农村部烟草生物学与加工重点实验室,山东省青岛市崂山区266101 [2]石河子大学机械电气工程学院,新疆维吾尔自治区石河子市832000 [3]中国烟草总公司重庆市公司,重庆市江北区400023 [4]中国烟草总公司重庆市公司彭水分公司,重庆市彭水县409600 [5]中国烟草总公司重庆市公司黔江分公司,重庆市黔江区409000

出　　处：《烟草科技》2024年第8期104-112,共9页Tobacco Science & Technology

基　　金：中国烟草总公司重庆市公司科技项目“烤烟图像识别智能烘烤技术研究与应用”(B20202NY1335);中国烟草总公司重点项目“基于图像精准识别的烟叶智能烘烤关键技术研究与应用”(110202102007)。

摘　　要：为解决烟叶烘烤过程中无法实时观察烟叶状态而导致烘烤工艺不匹配等问题,基于卷积神经网络(Convolutional Neural Network)和长短期记忆神经网络(Long Short-Term Memory)建立一种烟叶烘烤阶段识别CNN-LSTM模型。采用由工业相机和拍摄光源组成的图像采集装置采集烘烤过程中的烟叶图像数据,根据烟叶颜色、纹理变化将采集的图像数据划分为10个烘烤阶段,通过所建立的CNN-LSTM模型进行验证并与其他3种常用识别模型CNN、ResNet50和Inception-ResNet v2进行对比。结果表明:①CNN-LSTM模型对10个烘烤阶段训练集、验证集的识别准确率分别为99.55%和98.70%;对测试集的识别准确率和平均F1分数分别达到99.58%和99.36%,相比Inception-ResNet v2模型分别提高1.17和1.72百分点;②CNN-LSTM模型判断错误的样本远少于其他3种模型且大多为相邻烘烤阶段,从而有效保证烘烤工艺相匹配。该方法可为实现烟叶烘烤阶段快速和准确识别提供支持。In order to observe the status of tobacco leaves exactly in real time and set appropriate curing parameters accordingly in the process of flue-curing,a CNN-LSTM model for identifying tobacco flue-curing stages based on Convolutional Neural Network and Long Short-Term Memory Neural Network was developed.An image acquisition device composed of industrial cameras and light sources was used to collect tobacco image data during curing,and the acquired image data were categorized into 10 curing stages according to the color and texture changes of tobacco leaves.The CNN-LSTM model was verified and compared with three conventional recognition models:CNN,ResNet50 and Inception-ResNet v2.The results showed that:1)For the 10 flue-curing stages,the recognition accuracies of the training and verification sets of the CNN-LSTM model reached 99.55%and 98.70%,respectively.The recognition accuracy and average F1 score of the test set reached 99.58%and 99.36%,respectively,which were 1.17 and 1.72 percentage points higher than those of the Inception-ResNet v2 model.2)The number of samples which were incorrectly judged by the CNN-LSTM model was far less than that by the other three models,and most of them fell in adjacent curing stages.This method supports the rapid and accurate identification of tobacco flue-curing stages.

关 键 词：烟叶 烘烤阶段 卷积神经网络 长短期记忆神经网络 识别模型 准确率 

分 类 号：TS431[农业科学—烟草工业]