本文已被:浏览 445次 下载 203次
投稿时间:2022-08-16
投稿时间:2022-08-16
中文摘要: 针对白酒“探汽上甑”工艺在实现自动化过程中出现的探汽准确率低的问题,提出一种基于卷积神经网络的探汽方法。通过红外热成像仪采集甑锅内酒醅表面的红外图像并做预处理,再结合上甑工艺特点将图像分类,利用卷积神经网络训练得到探汽模型。训练结果表明,AlexNet、VGGNet-16、GoogLeNet、ResNet-18、DenseNet-37 的探汽准确率分别为0.997 0、0.998 0、0.994 2、0.989 8、0.997 0,综合考虑选用DenseNet-37 做模型评估,测试集测试的精确率为0.997 0,召回率为0.997 0,F1 分数为0.996 9,表示该模型性能表现好,故能满足探汽上甑要求。
Abstract:Aiming at the low accuracy of steam detection in the automation process of detecting stream on the retort of Chinese baijiu,a steam detection method based on convolutional neural network was proposed. The infrared images of the surface of fermented grains in the retort were collected by an infrared thermal imager and pre-processed, and then the images were classified based on the characteristics of the retorting process. The steam detection model was established after convolutional neural network training. The training results showed that the accuracy rates of AlexNet,VGGNet-16,GoogLeNet,ResNet-18,and DenseNet-37 were 0.997 0,0.998 0,0.994 2,0.989 8 and 0.997 0,respectively. Furthermore,DenseNet-37 was chosen for performance evaluation,which showed the precision rate of 0.997 0, the recall rate of 0.997 0,and the F1-score of 0.996 9.The results indicated that the DenseNet-37 model performed well and met the requirements of steam detection on the retort.
keywords: steam detection methods infrared images image pre - processing convolutional neural networks model evaluation
文章编号:202405019 中图分类号: 文献标志码:
基金项目:国家自然科学基金项目(42074218);四川省科技计划项目(2022YFS0554);四川省重大科技专项项目(2018GZDZX0045);四川省科技成果转移转化示范项目(2020ZHCG0040)
作者 | 单位 |
刘文斌1,2,庹先国1,2 *,张贵宇1,2,3 ,罗琪1,2,彭英杰1,2 | 1.四川轻化工大学 自动化与信息工程学院,四川 宜宾 644000;2.四川轻化工大学 人工智能四川省重点实验室,四川 宜宾 644000;3.西南科技大学 信息工程学院,四川 绵阳 621010 |
引用文本: