Identification Of Rice Geographical Origin By Spectroscopy Technology And Machine Learning Algorithms

Rice is one of the most important food crops in the world.Rice is rich in nutrition,providing carbohydrates,proteins,fats and a variety of mineral elements for the human body.Because the quality of rice is closely related to the external environment in which it grows,such as soil,sunshine and irrigation water,there are certain differences in the quality of rice produced by the same rice variety in different geographical areas.Laser-induced breakdown spectroscopy(LIBS)is a new substance analysis technique developed rapidly in recent years.It has incomparable advantages as other spectral techniques,such as simple sample preparation,on-line simultaneous detection of multiple elements,etc.It is widely used in element analysis and substance identification.Hyperspectral imaging(HSI)obtained twodimensional image information and one-dimensional spectral information of the target object at the same time by using the imaging spectrometer,which was characterized by spectral integration and high altitude spectral resolution.Due to the limitations of large amount of spectral data,high detection limit and low quantitative and qualitative accuracy of rice LIBS and HSI,machine learning algorithm was introduced to solve the problem of difficult processing of a large number of complex spectral data and improve the accuracy of qualitative measurement of spectral technology,so as to overcome the limitations of spectral technology.On this basis,this paper fuses the rice spectral data obtained by two spectral technologies,LIBS and HSI,compares them with the single spectral technology,and analyzes the differences between the single spectral data and the spectral fusion data used in the origin tracing of rice.The results showed that the combination of LIBS-HSI spectral data fusion and machine learning algorithm could quickly and accurately identify the rice origin.The work completed in this paper is as follows:1.For the LIBS and HSI experiments of rice,rice samples were processed into smooth rice samples and original hulled grains of the same size,respectively,and the parameters of the LIBS and HSI experiment systems were optimized.2.Noise reduction of the experimentally acquired spectra.Since the noise generated in the process of spectrum acquisition will affect the experimental results,according to the spectral characteristics of LIBS and HSI,wavelet transform and multiple scattering correction are selected to denoise LIBS and HSI spectra.3.Cluster analysis and classification of 12 kinds of rice from different producing areas were carried out.Principal component analysis(PCA)was used for the preliminary cluster analysis of rice,and then the rice spectrum was substituted into partial least squares discriminant analysis(PLS-DA),support vector machine(SVM)and extreme learning machine(ELM)for the identification of rice origin.LIBS and HSI two spectral data are in the highest accuracy of SVM,98.42% and 93.45%,respectively.4.Cluster analysis and classification identification of rice from two different producing areas were carried out based on spectral fusion data.Compared with the classification effect under single spectral data,the accuracy and stability of rice identification results under fusion data were higher.The accuracy of fusion data in SVM algorithm is up to 99.73%.