Research On The Detection Method Of Rice From Different Origins Based On Raman Spectroscopy

The detection and analysis of rice origin are particularly important for food quality and safety and for maintaining order in the food market.Traditional rice detection methods are mainly chemical detection and sensory detection,which have the disadvantages of cumbersome operation and strong subjectivity,and it is difficult to meet the requirements of fast,nondestructive,and portable detection.Raman spectroscopy has fingerprint characteristics.By analyzing the Raman spectral data of substances,the structural composition and chemical composition of substances can be analyzed.Its features such as simple operation,high sensitivity,and no need for marking provide the possibility for the classification,identification,and detection of rice from different origins.The convolutional neural network is a deep neural network based on convolution operation.It replaces the original artificial feature extraction process through the end-to-end feature extraction process,and can directly extract complex hidden features with multiple levels of abstraction from preprocessed data.Since the convolutional neural network can make full use of the characteristics of Raman spectroscopy,and has the advantages of processing complex spectral data,improving the accuracy of classification and recognition,it can be used for Raman spectroscopy analysis.The research of this topic uses Raman spectroscopy as a technical means,combined with convolutional neural network and Raman spectral information features,designs and establishes two rice classification models with different data types as input,the main contents are as follows:1.The development and current status of food safety,Raman spectroscopy analysis technology,machine learning,and convolutional neural network are introduced.In the current situation that the rice origin inspection and supervision system is confusing and the traditional rice classification and inspection methods have certain defects,it is necessary to propose a nondestructive and efficient rice origin inspection technology.The application of machine learning and convolutional neural networks in the field of spectroscopy is described,showing its good development and prospect in the field of spectroscopy.The principle of Raman spectroscopy,spectral pre-processing methods,and the theory of machine learning and deep learning algorithms are explained and specified to provide the theoretical basis for the experimental part.2.Data acquisition,preprocessing,spectral analysis,and data set preparation for rice Raman spectra,selection and construction of software and hardware environments for subsequent experiments.In this project,the Raman spectrometer was used to collect the Raman spectral information of rice from two different origins in the north and the south,and the collected Raman spectral signals were subjected to spectral preprocessing,characteristic peak analysis,and data set preparation of different sizes.Through spectral analysis,it can be seen that due to the high similarity of the spectral information characteristics of rice from different origins,it is impossible to detect and classify rice by traditional principal component analysis.This topic proposes a research method for establishing a high-accuracy classification model through deep learning combined with Raman spectroscopy.Then,according to the comparison of different deep learning environment frameworks,the hardware and software environment for subsequent modeling experiments was built.3.The rice Raman spectral classification model R-S-1D model and R-S-2D model based on the convolutional neural network were established.According to the characteristics of Raman spectral signals,two convolutional neural network models suitable for Raman spectral classification were designed with single-channel input and multi-channel input,and a comparative study was carried out on the two models by using the model evaluation index.The research shows that under the premise that the two models can accurately and quickly classify rice from different origins,the two models have different adaptation scenarios.In order to verify the advantages of the convolutional neural network in dealing with the identification of Raman spectra,the research of this subject established four machine learning models.Through comparison,it was proved that the convolutional neural network model has good performance in rice classification and identification.