Feature Extraction And Classification Of Near - Infrared Spectra Of Peanut Seed Quality

In this paper, three representative varieties of peanut seeds were selected for the experimentbased on visible/near-infrared reflectance spectroscopy living in the wavelength range from600to1100nm, and carried out the analysis of peanut seeds quality. Peanut seeds spectra were collected innatural light and tungsten halogen lamp as a light source. The spectral features of the original spectraldates were extracted by the wavelet analysis and the principal component analysis (PCA),then theMahalanobis distance discriminate model and Perceptron Neural Network were used for clusteranalysis and fast classification of spectral features. And wavelet analysis combining with principalcomponent regression was used to do rapid detection for the fat content of peanut seeds. The maincontents of the study include the following three aspects:(1) The spectral features of the visible/near-infrared reflectance spectral dates were extractedby the wavelet analysis.Take advantage of the characteristics of wavelet multi-resolution analysis,and make the pretreatment for the spectrum of the peanut seeds, so the outline of the main featuresand characteristics of the spectrum were extracted. The filter pretreatment of wavelet analysis caneffectively filter out the noise spectrum signals, highlight the weak spectral signals useful information,and establish qualitative and quantitative model for the data to provide a reliable source.(2) The recognition model of principal component analysis and variety of pattern recognitionwas established to identify the classification of peanut seeds.The principal component analysis (PCA)combining with Mahalanobis distance discriminate model and Perceptron Neural Network wereused for classification model of peanut seeds in this paper. Experimental results show that:Mahalanobis distance discriminant analysis model to predict peanut seed set have the discriminantaccuracy rate of95%; Perceptron neural network model to predict the set have the discriminantaccuracy rate93.33%.So two methods can distinguish peanut seed multi-target spectral identificationand have high spectral recognition rate, and provide a practical approach to distinguish the quality ofpeanut seeds.(3) The qualitative analysis model of the fat content of peanut seeds was established, and wasused for predict on the fat content of peanut seeds. The wavelet analysis combining with the principalcomponent regression (PCR) were used to predict on the fat content of peanut seeds. Achieved a higher prediction accuracy, prediction error was less than0.35, the relative error ranged from0.03%to1.03%. Simple experimental operation can meet the requirements of breeding problems in the rapidscreening of peanut seeds, and provide an effective approach to distinguish the quality of peanut seedsby taking advantage of the analysis of visible/near-infrared reflectance spectroscopy.