吕铷麟,贾镇,胡益滔,何洪源,何伟文.基于卷积神经网络的食品塑料包装袋光谱识别[J].包装工程,2022,43(3):121-128.
LYU Ru-lin,JIA Zhen,HU Yi-tao,HE Hong-yuan,HE Wei-wen.Spectral Recognition of Plastic Food Packaging Bags Based on Convolution Neural Network[J].Packaging Engineering,2022,43(3):121-128. |
| 基于卷积神经网络的食品塑料包装袋光谱识别 |
| Spectral Recognition of Plastic Food Packaging Bags Based on Convolution Neural Network |
| 投稿时间：2021-08-08 |
| DOI：10.19554/j.cnki.1001-3563.2022.03.015 |
| 中文关键词: 高光谱成像技术 卷积神经网络 包装袋 机器学习 快速识别 |
| 英文关键词: hyperspectral imaging technology convolutional neural network food plastic packaging bag machine learning fast identification |
| 基金项目:国家重点研发计划资助(2017YFC0822001) |
|
| 摘要点击次数: |
| 全文下载次数: |
| 中文摘要: |
| 目的 实现食品塑料包装袋的快速检测和材质区分。方法 研究使用高光谱成像技术在450～950 nm波长范围下采集了49组不同食品包装袋样本的光谱数据，利用Savitzky-Golay平滑滤波、数据归一化和主成分分析进行预处理，建立决策树、支持向量机2种传统机器学习模型和卷积神经网络模型，并比较了它们对包装袋材质的识别性能。结果 决策树模型与支持向量机模型的验证识别率分别为87.8%和88.9%，卷积神经网络模型的验证识别率高达100%，损失函数值最终下降到0.0171且达到收敛，在分类效果和精度上具有明显的优势。结论 高光谱检测方法不破坏检材，重现性好，稳定性强，实现了对食品塑料包装袋的精准识别。卷积神经网络模型对食品包装袋高光谱数据的识别效果最好，为食品包装袋质量检测领域中塑料包装袋的识别鉴定提供依据。 |
| 英文摘要: |
| The work aims to realize the rapid detection and material differentiation of food plastic packaging bags. The spectral data of 49 groups of different food packaging bags were collected by hyperspectral imaging technology in the wavelength range of 450～950 nm. The data were preprocessed by savitzky Golay smooth filtering, data normalization and principal component analysis to establish two traditional machine learning models of decision tree and SVM and one convolutional neural network model. Then, the recognition performance of traditional machine learning models and convolutional neural network model on the packaging bag materials was compared. The verification recognition rate of decision tree model and SVM model was 87.8% and 88.9%, respectively, while the verification recognition rate of convolutional neural network model was up to 100%, and the loss function value finally dropped to 0.0171 and tended to be stable. Therefore, the convolutional neural network model had obvious advantages in classification effect and accuracy. The method of hyperspectral detection does not destroy the material, and has good reproducibility and strong stability, which can realize the accurate identification of food plastic packaging bags. The convolutional neural network model has the best recognition effect on hyperspectral data of food packaging bags and provides the basis for the identification and recognition of plastic packaging bags in the field of food packaging quality detection. |
|
查看全文
查看/发表评论 下载PDF阅读器 |
| 关闭 |
渝公网安备 50010702501716号