摘要
烤烟的等级识别是烟草产业的一个关键环节,为了降低烟农烤烟分级劳动强度,减少主观因素,提高识别精度,需要实现烤烟的自动分级。研究了深度学习中多种卷积神经网络的多层特征提取方法,并基于ShuffleNetV2网络提出一种改进的网络模型(ShuffleNetV2_FTC)。ShuffleNetV2_FTC网络是将ShuffleNetV2网络的主干单元进行更改,并引入CBAM(Convolutional Block Attention Module)注意力机制和SiLU激活函数。应用该模型对27种类别的烤烟图像进行识别分类。该模型的最佳测试准确率为93.09%,检测帧率达到每秒15.3张。相对原模型获得了0.24%(0.5×)、6.06%(1×)和4.73%(1.5×)的准确率提升,每秒检测12.3张图片到15.3张图片的检测帧率提升。ShuffleNetV2_FTC网络结合机器视觉技术可以很好地识别烤烟等级,为优化烤烟的收购、烘干和加工等流程奠定基础。
The grade identification of flue-cured tobacco is a crucial step in the tobacco industry.In order to reduce the labor intensity of tobacco farmers in grading flue-cured tobacco,minimize subjective factors,and enhance identification accuracy,it is necessary to achieve the automatic grading of flue-cured tobacco.This article investigated multiple methods of multi-layer feature extraction using various convolutional neural networks in deep learning.Based on ShuffleNetV2,an improved model(ShuffleNetV2_FTC)was proposed.The ShuffleNetV2_FTC modified the backbone units of ShuffleNetV2 and incorporated the CBAM(Convolutional Block Attention Module)attention mechanism and SiLU activation function.This model was applied to the identification and classification of 27 categories of flue-cured tobacco images.The testing accuracy of this model reached 93.2%,and the detection frame rate achieved 15.3 frames per second.Compared to the original model,there was an improvement of 0.24%(0.5×),6.06%(1×)and 4.73%(1.5×)in accuracy,as well as an increase in the detection frame rate from 12.3 frames per second to 15.3 frames per second.The ShuffleNetV2_FTC,combined with machine vision technology,can effectively identify the grades of flue-cured tobacco,laying the foundation for optimizing the procurement,drying and processing processes of flue-cured tobacco.
作者
冯川
祝诗平
黄华
严森垚
于丽敏
FENG Chuan;ZHU Shiping;HUANG Hua;YAN Senyao;YU Limin(College of Engineering and Technology,Southwest University,Chongqing 400715,China;School of Information Science and Engineering,Shandong Agriculture and Engineering University,Jinan 250110,China)
出处
《西南大学学报（自然科学版）》
CAS
北大核心
2025年第1期213-225,共13页
Journal of Southwest University(Natural Science Edition)
基金
教育部中国高校产学研创新基金—德州专项(2021DZ005)
中国烟草总公司云南省公司科技计划项目(2021530000241036)。
