烟草科技2026，Vol.59Issue(9)：63-73,11.DOI:10.16135/j.issn1002‒0861.2025.0480
基于MAFF-CNN-LSTM的雪茄烟叶晾制阶段识别方法
Identification method of air-curing stages of cigar tobacco leaves using MAFF-CNN-LSTM
摘要
Abstract
Manual visual assessment of the air-curing status of cigar tobacco leaves is highly experience-dependent and may lead to inaccurate stage identification.To address this issue,a MAFF-CNN-LSTM model was developed for identifying the air-curing stage of cigar tobacco leaves.The model integrates a multi-region adaptive feature fusion architecture with a convolutional neural network and long short-term memory networks.Images of cigar tobacco leaves were collected during the air-curing process using an image acquisition system.The image data were classified into five air-curing stages according to changes in leaf status.Sample diversity was increased using conventional data augmentation and Mixup-based augmentation.The proposed model was compared with CNN-LSTM,MAFF-CNN,CNN and ResNet50 models to evaluate its effectiveness.The results showed that:1)MAFF-CNN-LSTM model achieved recognition accuracies of 98.72%and 97.45%on the training and validation sets respectively for the five air-curing stages.On the test set,the model achieved an accuracy of 98.29%and an average F1 score of 98.22%.Compared with the CNN-LSTM and MAFF-CNN models,the proposed model improved the test accuracy by 4.40 and 1.53 percentage points respectively.2)Moreover,the MAFF-CNN-LSTM model produced far fewer misclassified samples than the other four models,and the misclassifications occurred only between adjacent air-curing stages.These results indicate that the proposed model can provide a reliable basis for precise regulation of temperature and humidity parameters at different air-curing stages.The model supports rapid and accurate identification of air-curing stages of cigar tobacco leaves and contributes to digital management of the cigar tobacco air-curing process.关键词
雪茄烟叶/晾制阶段/多区域自适应特征融合/卷积神经网络/长短期记忆神经网络/阶段识别Key words
Cigar tobacco leaf/Air-curing stage/Multi-region adaptive feature fusion/Convolutional neural network/Long short-term memory neural network/Stage identification分类
轻工纺织引用本文复制引用
顾瑞,朱荣光,孟令峰,江桂花,黄中涛,孟霖,常鹏飞..基于MAFF-CNN-LSTM的雪茄烟叶晾制阶段识别方法[J].烟草科技,2026,59(9):63-73,11.基金项目
福建省烟草公司龙岩市公司科技项目"雪茄烟叶晾制进程中异常烟叶预警模型的构建与应用"(2024350825240021). （2024350825240021）
