烟草科技2025，Vol.58Issue(10)：33-46,14.DOI:10.16135/j.issn1002-0861.2024.0883
基于EfficientNet-GECA模型的烘烤过程关键温度点烟叶状态识别
Recognition of tobacco leaf status at key temperature points during flue-curing process based on EfficientNet-GECA model
摘要
Abstract
In order to intelligently monitor the status of tobacco leaves during flue-curing process,image acquisition devices were used to collect image data of tobacco leaves,and an recognition model for the yellowing and drying status of tobacco leaves was developed by using manual annotation and an improved EfficientNet-GECA model based on EfficientNet-B0 to analyze the changes of tobacco leaves at key temperature points during 354 flue-curing processes in three tobacco producing areas including Nanyang,Sanmenxia,and Pingdingshan in Henan Province.The results showed that:1)Compared with the classic neural network models such as MobileNetV2,MobileNetV3,VGG16,ShuffleNetV2,ResNet50 and EfficientNet-B0,the EfficientNet-GECA model increased the recognition accuracies of the test set for the yellowing and drying status of tobacco leaves by 1.81 to 32.36 percentage points and 1.98 to 25.84 percentage points,respectively,with recognition accuracies of 88.74%and 80.47%,respectively.2)The main yellowing status of tobacco leaves from Nanyang,Sanmenxia and Pingdingshan production areas at the key temperature points of 38,40,42,45,48 and 54℃during the flue-curing process were relatively consistent,while there were significant differences in the drying status of tobacco leaves among the different production areas at the key temperature points including 40,42 and 54℃.The established recognition model for tobacco leaf yellowing and drying status based on EfficientNet-GECA model can be used for intelligent monitoring of tobacco leaf status during flue-curing process,providing a basis for developing customized regulation and control strategies during tobacco leaf curing process.关键词
烟叶烘烤/图像数据集/深度学习/变黄状态/干燥状态Key words
Tobacco flue-curing/Image dataset/Deep learning/Yellowing status/Drying status分类
农业科技引用本文复制引用
张恒,陈栋,张富生,徐嫱,陈广晴,过伟民,周硕野,成钊,蔡宪杰,刘剑君,李俊营,张艳玲,王爱国..基于EfficientNet-GECA模型的烘烤过程关键温度点烟叶状态识别[J].烟草科技,2025,58(10):33-46,14.基金项目
中国烟草总公司重大科技项目"烟叶高质量科学图像数据集构建关键技术研究"(110202201051) （110202201051）
河南省烟草公司重点科技项目"基于图像识别技术的热泵烤房智能烘烤技术研究与应用"(2024410000240029) （2024410000240029）
河南省烟草公司三门峡市公司科技项目"基于数字孪生技术的烟叶智能烘烤工场关键技术研究与应用"(2024411200200019X). （2024411200200019X）
