Study On Type Identification Of Rice Seeds Based On Laser Induced Breakdown Spectroscopy

For a long time,the phenomenon of fake and inferior seeds in rice seed market has emerged in endlessly,resulting in serious damage to farmers’ interests and market economy.The main reason is that it is difficult to classify and identify rice seed varieties.At present,the commonly used seed identification methods have low identification accuracy and efficiency,which requires a lot of time and manpower.Therefore,it is necessary to seek a fast and accurate identification method for batch detection of rice seeds.Laser-induced breakdown spectroscopy(LIBS)is a new material identification method in recent years,in view of the great development potential of LIBS technology,this paper studied the rapid identification of rice seed varieties based on LIBS technology and BP(Back propagation)neural network technology.In this paper,11 segments of LIBS spectra of rice seeds from 10 different regions were collected.The full spectrum of rice seeds was obtained by synthesizing the 11 segments of LIBS spectra,and the characteristic spectra of K,Si,Mg,Ca and Na elements in rice seeds were extracted at the same time.Next,the spectral line integration method,the discrete inverse Fourier transform method,the discrete wavelet transform method and the intensity ratio method were used to preprocess the samples’ spectra,respectively.The full spectrum,segmented spectrum and characteristic spectrum identification models of rice seeds were established by BP neural network,and the spectra before and after pretreatment were identified and analyzed.After preprocessing of full spectrum and segmented spectrum by spectral line integration method and discrete wavelet transform method,the identification accuracy of full spectrum and segmented spectrum with a central wavelength of 405 nm were improved by about 6%～10%,reaching more than 80%.But the speed of full spectrum identification were slow,and the identification time were more than 160 seconds.After the preprocessing by discrete inverse Fourier transform,the accuracy of the full-spectrum identification model and the segmented-spectrum identification model with center wavelength of 405 nm increased by about 22%,reaching 98.27% and98.04%,respectively;At the same time,their identification time were greatly reduced to20.22 seconds and 14.79 seconds,respectively.For the characteristic spectral model,after preprocessing by intensity ratio method and discrete inverse Fourier transform method,the identification effect and identification speed all decreased;After processing by discrete wavelet transform,the identification effect of characteristic spectrum was slightly improved,the identification accuracy was 96.53% and the identification time was 8.76 seconds.The research results show that the time-domain full spectrum identification model and the time-domain segmented-spectrum identification model with a central wavelength of 405 nm after discrete inverse Fourier transform,and the characteristic spectrum identification model after discrete wavelet transform are the best;Their classification accuracy for 10 varieties of rice seed samples exceeded 96%,and the identification time were only need a few seconds.In addition,the relative standard deviation of the identification accuracy of the above three models are less than 2%,indicating that the three identification models have strong stability and repeatability,and the identification results are accurate and reliable,which can provide a new method for rapid and efficient identification of rice varieties.On the other hand,inverse Fourier transform and wavelet transform as spectral preprocessing methods have certain application potential in the classification and identification research of LIBS spectra,which can provide new ideas for the identification of other substances.