基于1D-DenseRNet的罐头真空度数据分类检测方法研究

doi:10.3969/j.issn.1004-1524.20240092

浙江农业学报 ›› 2024, Vol. 36 ›› Issue (12): 2846-2856.DOI: 10.3969/j.issn.1004-1524.20240092

基于1D-DenseRNet的罐头真空度数据分类检测方法研究

俞烁辰¹(), 张俊²^,^*(), 宋新杰¹^,³, 周锦云²

1.浙江科技大学生物与化学工程学院,浙江杭州 310023
2.浙江省农业科学院食品科学研究所,农业农村部果品采后处理重点实验室,浙江杭州 310021
3.浙江省农产品化学与生物加工技术重点实验室,浙江杭州 310023

收稿日期:2024-01-21 出版日期:2024-12-25 发布日期:2024-12-27
作者简介:俞烁辰(1999—),男,江苏无锡人,硕士研究生,研究方向为智能制造技术、深度学习。E-mail:1144140917@qq.com
通讯作者: *张俊,E-mail: hunterzju@163.com
基金资助:
国家现代农业产业技术体系(CARS-26-04BY)

Research on 1D-DenseRNet-based classification and detection method for canned food vacuum data

YU Shuochen¹(), ZHANG Jun²^,^*(), SONG Xinjie¹^,³, ZHOU Jinyun²

1. College of Biological and Chemical Engineering, Zhejiang University of Science and Technology, Hangzhou 310023, China
2. Key Laboratory of Fruit Postharvest Processing, Ministry of Agriculture and Rural Affairs, Institute of Food Science, Zhejiang Academy of Agricultural Sciences, Hangzhou 310021, China
3. Key Laboratory of Chemistry and Bioprocessing Technology for Agricultural Products in Zhejiang Province, Hangzhou 310023, China

Received:2024-01-21 Online:2024-12-25 Published:2024-12-27

摘要/Abstract

摘要：

针对食品罐头真空度检测精度低、成本高的问题,为实现食品罐头真空度的无损检测,提高检测速度与精确度,提出利用激光位移传感器测量罐头顶部曲线数据,通过1D-DenseRNet模型对曲线数据进行分类的研究。该模型包括多个改进的密集连接模块、门循环单元、注意力层和残差连接模块,提取时序电压序列特征和捕捉长期依赖关系。采用交叉验证的方法,分析了不同网络层对模型性能的影响。通过改变卷积层、结合注意力机制、不同的循环神经网络模块和残差网络结构,观察模型的准确率、模型大小等其他评价指标的变化,设计出最优的网络模型结构。实验数据表明,动量因子设定为0.9,学习率设定为0.000 5时,结合了门循环单元的1D-DenseRNet模型小样本数据集上达到了最高的准确率(98.77%),且模型参数量也相对较小。对比单一的卷积神经网络及其他混合网络,展现了1D-DenseRNet模型处理食品罐头真空度检测任务的优势。

关键词: 罐头无损检测, 神经网络, 真空度检测, 小样本数据分类

Abstract:

Aiming at the problem of low accuracy and high cost of vacuum detection in food cans, in order to realize the nondestructive detection of vacuum in food cans and improve the detection speed and accuracy, the research of measuring the curve data on the top of cans by using laser displacement sensors and categorizing the curve data by 1D-DenseRNet model is proposed. The model includes several improved dense connectivity modules, gate cycle units, attention layers and residual connectivity modules to extract time-series voltage sequence features and capture long-term dependencies. A cross-validation approach was used to analyze the effect of different network layers on the model performance. The optimal network model structure is designed by changing the convolutional layer, combining the attention mechanism, different recurrent neural network modules and the residual network structure, and observing the changes in other evaluation indexes such as model accuracy and model size. The experimental data show that the 1D-DenseRNet model incorporating the gate recurrent unit achieves the highest accuracy (98.77%) on the small sample dataset when the momentum factor is set to 0.9 and the learning rate is set to 0.000 5, and the number of model parameters is also relatively small. Comparison with a single convolutional neural network and other hybrid networks demonstrates the advantages of the 1D-DenseRNet model in handling the task of vacuum detection in food cans.

Key words: non-destructive testing of food cans, neural network, vacuum detection, small sample data classification

中图分类号:

TS205.6

俞烁辰, 张俊, 宋新杰, 周锦云. 基于1D-DenseRNet的罐头真空度数据分类检测方法研究[J]. 浙江农业学报, 2024, 36(12): 2846-2856.

YU Shuochen, ZHANG Jun, SONG Xinjie, ZHOU Jinyun. Research on 1D-DenseRNet-based classification and detection method for canned food vacuum data[J]. Acta Agriculturae Zhejiangensis, 2024, 36(12): 2846-2856.

图/表 14

参考文献 20

[1]	曾颖, 莫锡乾. 我国罐头行业质量安全问题研究分析[J]. 现代食品, 2019, 25(10):128-130.
	ZENG Y, MO X Q. Research and analysis on quality and safety of canning industry in China[J]. Modern Food, 2019, 25(10):128-130. (in Chinese with English abstract)
[2]	仇凯, 邵懿, 王亚, 等. 罐头食品安全标准体系国内外对比分析研究[J]. 中国食品卫生杂志, 2021, 33(4):509-517.
	QIU K, SHAO Y, WANG Y, et al. Comparative analysis and research on safety index of domestic and international canned food standard[J]. Chinese Journal of Food Hygiene, 2021, 33(4):509-517. (in Chinese with English abstract)
[3]	陆新龙. 罐头的“打检” 技术[J]. 江苏食品与发酵, 1992(2):20-21, 16.
	LU X L. The “inspection” technology for canned goods[J]. Jiangsu Food and Fermentation, 1992(2):20-21, 16. (in Chinese)
[4]	干蜀毅, 朱武, 陈长琦, 等. 电涡流法检测罐头真空度的探头研制和仪器智能化[J]. 真空科学与技术, 2002, 22(5):326-328.
	GAN S Y, ZHU W, CHEN C Q, et al. Non-destructive detection of low vacuum inside canned food and its sensors[J]. Chinese Journal of Vacuum Science and Technology, 2002, 22(5):326-328. (in Chinese with English abstract)
[5]	李莉. 双探头电涡流传感器真空度无损检测系统研究[D]. 天津: 天津大学, 2007.
	LI L. Research on non destructive testing system for vacuum degree of dual probe eddy current sensor[D]. Tianjin:Tianjin University, 2007. (in Chinese with English abstract)
[6]	韩威, 周松斌, 刘忆森, 等. 基于麦克风阵列的罐装食品真空度在线检测[J]. 中国测试, 2019, 45(7):128-133.
	HAN W, ZHOU S B, LIU Y S, et al. Online detection for vacuum of canned food based on microphone array[J]. China Measurement & Test, 2019, 45(7):128-133. (in Chinese with English abstract)
[7]	LECUN Y, BOTTOU L, BENGIO Y, et al. Gradient-based learning applied to document recognition[J]. Proceedings of the IEEE, 1998, 86(11):2278-2324.
[8]	严春满, 王铖. 卷积神经网络模型发展及应用[J]. 计算机科学与探索, 2021, 15(1):27-46.
	YAN C M, WANG C. Development and application of convolutional neural network model[J]. Journal of Frontiers of Computer Science and Technology, 2021, 15(1):27-46. (in Chinese with English abstract)
[9]	EREN L, INCE T, KIRANYAZ S. A generic intelligent bearing fault diagnosis system using compact adaptive 1D CNN classifier[J]. Journal of Signal Processing Systems, 2019, 91(2):179-189.
[10]	YU Z J, LU Y Q, AN Q, et al. Real-time multiple gesture recognition:application of a lightweight individualized 1D CNN model to an edge computing system[J]. IEEE Transactions on Neural Systems and Rehabilitation Engineering, 2022, 30:990-998.
[11]	JANA G C, SHARMA R, AGRAWAL A. A 1D-CNN-spectrogram based approach for seizure detection from EEG signal[J]. Procedia Computer Science, 2020, 167:403-412.
[12]	KIRANYAZ S, INCE T, GABBOUJ M. Real-time patient-specific ECG classification by 1-D convolutional neural networks[J]. IEEE Transactions on Bio-Medical Engineering, 2016, 63(3):664-675.
[13]	李文. 神经网络中梯度消失的解决办法[J]. 电脑知识与技术, 2023, 19(10):19-21, 28.
	LI W. Solution to vanishing gradients in neural networks[J]. Computer Knowledge and Technology, 2023, 19(10):19-21, 28. (in Chinese)
[14]	HE K M, ZHANG X Y, REN S Q, et al. Deep residual learning for image recognition[C]// 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). June 27-30, 2016, Las Vegas, NV, USA. IEEE, 2016:770-778.
[15]	HUANG G, LIU Z, VAN DER MAATEN L, et al. Densely connected convolutional networks[C]// 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). July 21-26, 2017, Honolulu, HI, USA. IEEE, 2017:2261-2269.
[16]	CHUNG J, GULCEHRE C, CHO K H, et al. Empirical evaluation of gated recurrent neural networks on sequence modeling[J]. Eprint Arxiv, 2014.DOI:10.48550/arXiv.1412.3555.
[17]	ZHANG J J, LIU Y H, YUAN H. Attention-based residual BiLSTM networks for human activity recognition[J]. IEEE Access, 2023, 11:94173-94187.
[18]	JOHN A, CARDIFF B, JOHN D. A 1D-CNN based deep learning technique for sleep apnea detection in IoT sensors[C]// 2021 IEEE International Symposium on Circuits and Systems (ISCAS). May 22-28, 2021, Daegu, Korea. IEEE, 2021:1-5.
[19]	TANG S N, ZHU Y, YUAN S Q. An improved convolutional neural network with an adaptable learning rate towards multi-signal fault diagnosis of hydraulic piston pump[J]. Advanced Engineering Informatics, 2021, 50:101406.
[20]	WEN L, GAO L, LI X Y, et al. Convolutional neural network with automatic learning rate scheduler for fault classification[J]. IEEE Transactions on Instrumentation and Measurement, 2021, 70:3509912.

真空度等级 Vacuum level	训练集 Training set	测试集 Test set	验证集 Valid set	合计 Total
0级(标准大气压)	3 645	729	729	5 103
Class 0 (standard atmospheric pressure)
1级(低真空度)	3 645	729	729	5 103
Class 1 (low vacuum)
合计Total	7 290	1 458	1 458	10 206

真空度等级 Vacuum level	训练集 Training set	测试集 Test set	验证集 Valid set	合计 Total
0级(标准大气压)	3 645	729	729	5 103
Class 0 (standard atmospheric pressure)
1级(低真空度)	3 645	729	729	5 103
Class 1 (low vacuum)
合计Total	7 290	1 458	1 458	10 206

结构名称 Structure name		超参数设置 Hyperparameter setting		训练超参数 Training hyperparameters
Dense	Dense	1DCNN	Kernel-size	12
block	layer		K-Stride	1
		1DCNN	Kernel-size	1
			K-Stride	1
		1DCNN	Kernel-size	3
			K-Stride	1
	Transition	1DCNN	Kernel-size	1
	layer	MaxPooling1D	Pool-Size	2
其他参数		优化算法Optimization algorithm		Nadam
Other parameters		迭代次数Number of iterations		100
		学习率Learning rate		0.000 5
		激活函数Activation function		Swish
		批次数据量Batch data volume		32
		损失函数Loss function		交叉损失函数
				Cross-loss function

结构名称 Structure name		超参数设置 Hyperparameter setting		训练超参数 Training hyperparameters
Dense	Dense	1DCNN	Kernel-size	12
block	layer		K-Stride	1
		1DCNN	Kernel-size	1
			K-Stride	1
		1DCNN	Kernel-size	3
			K-Stride	1
	Transition	1DCNN	Kernel-size	1
	layer	MaxPooling1D	Pool-Size	2
其他参数		优化算法Optimization algorithm		Nadam
Other parameters		迭代次数Number of iterations		100
		学习率Learning rate		0.000 5
		激活函数Activation function		Swish
		批次数据量Batch data volume		32
		损失函数Loss function		交叉损失函数
				Cross-loss function

学习率 Learning rate	准确率 Accuracy	精确度 Precision	F1值 F1 score	召回率 Recall
0.05	75.31 c	79.78 c	79.01 c	79.32 c
0.000 5	98.77 a	98.81 a	98.77 a	98.77 a
0.000 1	79.01 b	83.66 b	81.95 b	81.48 b

基于1D-DenseRNet的罐头真空度数据分类检测方法研究

Research on 1D-DenseRNet-based classification and detection method for canned food vacuum data

RichHTML

PDF (PC)

可视化

摘要/Abstract

引用本文

使用本文

图/表 14

参考文献 20

相关文章 15

编辑推荐

Metrics

本文评价

卷积核参数 Convolutional kernel parameters	准确率 Accuracy	精准度 Precision	F1值 F1 score	召回率 Recall
1	88.89 d	89.99 d	89.09 d	88.89 d
3	97.51 b	97.70 b	97.55 b	97.53 b
12	92.59 c	90.91 c	92.06 c	94.44 c
梯度设计	98.77 a	98.81 a	98.62 a	99.07 a
Gradient design

模型名称 Model name	准确率 Accuracy/%	精确度 Precision/%	F1值 F1 score/%	召回率 Recall/%	网络参数量 Number of network parameters/bytes
1D-CNN	48.14 g	30.51 i	36.58 i	45.68 h	687 213 c
1D-DenseNet	92.59 f	92.99 h	92.65 g	92.59 f	254 568 f
1D-DenseNet-LSTM	96.30 d	96.47 e	96.29 e	96.30 d	428 334 d
1D-DenseNet-Res-LSTM	97.53 c	97.62 c	97.51 c	97.53 c	428 334 d
1D-DenseNet-BiLSTM	95.31 e	97.70 b	97.53 b	97.55 b	924 846 a
1D-DenseNet-Res-BiLSTM	96.30 d	96.67 d	96.33 d	96.30 d	700 590 b
1D-DenseNet-GRU	97.83 b	94.37 g	93.85 f	93.83 e	386 286 e
1D-Swin Transformer	92.59 f	95.24 f	80.00 h	86.96 g	108 482 g
1D-DenseRNet	98.77 a	98.81 a	98.62 a	99.07 a	386 286 e

[1]	程陈, 董朝阳, 郑生宏, 周宇博, 钟宁, 李文明, 朱阳春, 丁枫华, 冯利平, 黎贞发. 光温因子驱动的园艺作物叶龄模型模拟精度比较[J]. 浙江农业学报, 2024, 36(6): 1368-1378.
[2]	朱冬琴, 冯全, 张建华. 基于剪枝的植物病害识别方法[J]. 浙江农业学报, 2023, 35(6): 1462-1472.
[3]	樊闯, 赵子皓, 张雪松, 杨沈斌. 基于BP神经网络的一季稻发育期预测模型[J]. 浙江农业学报, 2023, 35(2): 434-444.
[4]	柴琴琴, 曾建, 张勋. 基于贝叶斯优化卷积神经网络的金线莲伪品鉴别[J]. 浙江农业学报, 2022, 34(2): 391-396.
[5]	李超, 李锋, 黄炜嘉. 基于并联卷积神经网络的水果品种识别[J]. 浙江农业学报, 2022, 34(11): 2533-2541.
[6]	吉训生, 姜晓卫, 夏圣奎. 基于LSTM-Kalman模型的蛋鸡产蛋率预测方法[J]. 浙江农业学报, 2021, 33(9): 1730-1739.
[7]	张晴晴, 刘连忠, 宁井铭, 吴国栋, 江朝晖, 李孟杰, 李栋梁. 基于YOLOV3优化模型的复杂场景下茶树嫩芽识别[J]. 浙江农业学报, 2021, 33(9): 1740-1747.
[8]	冼炜轩, 尚国琲, 刘巧芹, 刘玉. 基于神经网络法与加权模型的乡村旅游用地竞争力评价研究——以北京市密云区为例[J]. 浙江农业学报, 2021, 33(8): 1519-1528.
[9]	杨红云, 罗建军, 孙爱珍, 万颖, 易文龙. 基于图像特征的水稻叶片全氮含量估测模型研究[J]. 浙江农业学报, 2020, 32(12): 2232-2243.
[10]	鲍烈, 王曼韬, 刘江川, 文波, 明月. 基于卷积神经网络的小麦产量预估方法[J]. 浙江农业学报, 2020, 32(12): 2244-2252.
[11]	王枭轩, 孟庆岩, 张海香, 魏香琴, 杨泽楠. 基于粒子群神经网络模型反演玉米、小麦叶面积指数[J]. 浙江农业学报, 2019, 31(7): 1170-1176.
[12]	蒋丰千, 李旸, 余大为, 孙敏, 张恩宝. 基于Caffe卷积神经网络的大豆病害检测系统[J]. 浙江农业学报, 2019, 31(7): 1177-1183.
[13]	李颀, 赵洁, 杨柳, 王俊, 高一星. 基于GA-BP神经网络和特征向量优化组合的黄瓜叶片病斑识别[J]. 浙江农业学报, 2019, 31(3): 487-495.
[14]	金秀, 卢杰, 傅运之, 王帅, 许高健, 李绍稳. 基于深度卷积神经网络的小麦赤霉病高光谱病症点分类方法[J]. 浙江农业学报, 2019, 31(2): 315-325.
[15]	杨国亮, 许楠, 康乐乐, 龚曼, 洪志阳. 基于参数指数非线性残差神经网络的脐橙病变叶片识别[J]. 浙江农业学报, 2018, 30(6): 1073-1081.