Analysis Of Cotton Internal Quality Index Based On Spectrum

The content of cotton quality inspection is gradually changed from the traditional grade,length,moisture and impurity to the physical index reflecting the inner quality of the cotton: the value of the micronaire and the index of strength.Traditional methods of measuring indicators can be arduous,and artificial error is not very easy to be avoided.The large capacity tester that can measure cotton quality accurately and quickly is expensive,which is not conducive to popularization.Therefore,a method based on spectral detection for physical quality of cotton inner quality is studied in this paper.This method mainly uses near-infrared and mid-infrared spectral data,combined with the standard data obtained by large capacity tester to do modeling analysis.The specific contents are as follows:(1)Establishing spectral experiments and quantitative analysis models.Near-infrared and mid-infrared spectra of samples were collected by Fourier near-infrared spectroscopy and Brook infrared spectrometer.A quantitative model of partial least square method(PLS)and partial least squares combined support vector machine method(PLS-SVM)are established in the whole spectrum area of near-infrared and middle-infrared.The results of the model are evaluated by five indexes,which are the predicted root-mean-square error,the relative predicted root-mean-square error,the determination system number,the verifying root-mean-square error and the verification set coefficient.The results show that the prediction effect of the model is better after the smoothing treatment,and the model predicted by PLS-SVM is better than the PLS.(2)Establish a qualitative model after wavelength screening.In order to further improve the accuracy and the efficiency of model prediction,and to carried out near-infrared and middle-infrared spectra by using non-information variable elimination method and non-information variable elimination combined with continuous projection method,this paper proposed two-dimensional correlation spectroscopy to select the wavelength of near-infrared spectroscopy,and establishing a quantitative model by using PLS-SVM method.The results show that in thenear-infrared spectrum area,the prediction effect of the established model after two kinds of wavelengths filtering are improved,but in the middle-infrared spectrum area,they were both reduced.Therefore,we proposed a wavelength selection method based on Gaussian software and absorbance is to filter the mid-infrared spectrum,and PLS-SVM algorithm is used to verify its performance.(3)Set up a classification model.In order to improve the classification performance of the micronaire,the Fisher discriminant algorithm and three different forms of self organizing mapping classification combined by the results of the optimal band were used to classify the cotton’s micronaire.The results show that the classification performance of the Fisher discriminant algorithm is poor,and although the supervised self organizing mapping algorithm has the best classification performance,only a limited number of labeled samples can be uesd is unable to be ignored.But the semi supervised self-organizing mapping algorithm can solve this problem and it’s classification performance is better.