摘要
根据不同产地大豆成分特征含量的差异,提出了一种基于电子舌结合元学习(meta-learning)-卷积神经网络(convolution neural networks,CNN)组合模型实现对大豆产地溯源的快速检测的方法。采用一维卷积神经网络对电子舌信号进行特征提取和分类识别,针对CNN模型难以适应新任务,依赖大量数据训练等问题,采用模型无关元学习算法(model-agnostic meta-learning,MAML)在预训练数据集上对CNN进行预训练,为CNN获得一个全局最优初始化参数。在此基础上,利用微调策略实现对新类别少量样本的快速适应与学习,最后通过模型实现查询样本的分类预测。实验结果表明,模型测试的准确率、召回率、精确率、F1-Score分别达到93.6%、93.8%、93.6%、0.937。研究为大豆产地溯源检测提供了一种快速的检测方法,并为仿生智能感官技术在农产品检测领域提供新的研究思路。
According to the differences of soybean component characteristics in different areas,this paper proposed a rapid detection method based on electronic tongue combined with meta-learning and convolution neural networks(CNN)combined model to realize soybean origin tracing.One-dimensional convolutional neural network is used to extract features and classify and recognize electronic tongue signals.Aiming at the problems that CNN model is difficult to adapt to new tasks and relies on a large amount of data training,model-agnostic meta-learning(MAML)algorithm is used to pre-train CNN on the pre-training dataset.Obtain a global optimal initialization parameter for CNN.On this basis,the fine-tuning strategy is used to quickly adapt and learn a small number of samples of the new category.Finally,the model is used to realize the classification prediction of query samples.The experimental results show that the accuracy,recall,precision and F1-Score of the model test are 93.6%,93.8%,93.6%and 0.937,respectively.This study provides a rapid detection method for soybean origin tracing and provides a new research idea for bionic intelligent sensory technology in the field of agricultural products detection.
作者
陈立同
高文
金鑫宁
张擎
王志强
Chen Litong;Gao Wen;Jin Xinning;Zhang Qing;Wang Zhiqiang(School of Computer Science and Technology,Shandong University of Technology,Zibo 255049,China;Zibo Industrial Digital Economy Development Center,Zibo 255049,China)
出处
《国外电子测量技术》
北大核心
2022年第12期140-147,共8页
Foreign Electronic Measurement Technology
基金
山东省自然科学基金(ZR2019MF024)
教育部科技发展中心产学研创新基金(2018A02010)项目资助。
关键词
大豆
产地溯源
电子舌
元学习
卷积神经网络
soybean
origin traceability
electronic tongue
meta learning
convolutional neural network
