Study On Ripeness Identification Of Banana Based On Optical Imaging And Spectral Analysis

Banana is one of the main fruits in subtropical and tropical regions.However,determination of bananaripeness degree detection technology of banana lags behind the development needs of its modern industry.Therefore,it is of great significance to study how to determine the ripeness of banana quickly and nondestructively for banana quality classification.In this paper,banana ripeness was studied from optical imaging and spectral analysis,which provided a reliable basis for nondestructive testing of banana ripeness and the development of nondestructive testing of the whole fruit industry.The main research contents and results are as follows:(1)Banana ripeness was distinguished based on image features.Bananas for destructive test to measure the total Soluble Solids(SSC)Content,the result shows that this experiment of artificial samples have reliability ripeness level.The threshold processing of r-b diagram of banana samples was carried out by Otsu method,and the foreground target was segmented.Then,the boxplot of H,a*,B *,c* and H * color components of banana samples were analyzed.The results showed that the color component values of H and a*were significantly different in bananas of different ripeness.Based on the methods of Support Vector Machine(SVM),linear discriminant analysis(LDA)and Naive Bayes(NB),the discriminant model was established based on the features of the color features,texture features and the Histogram of Oriented Gradient(HOG)of banana samples.The results showed that,The accuracy of banana ripeness discriminant model based on color feature is98.15%.(2)A banana ripeness discriminant model was established based on hyperspectral imaging technology.Compare the pretreatment effects of four methods combined with SVM discriminant model: Multiplicative Scatter Correction(MSC),Standard Normal Variate(SNV),savitzky-golay Smoothing(SGS),first-order differential and Procrustes Analysis(PA)on the spectral data of banana samples,and compare different spectral pretreatment methods.The results show that MSC and PA pretreatment make the scattered spectral curves relatively concentrated,reduce the intra-class variance of the data,and solve the problem of spectral baseline drift.The model after pretreatment has a good effect,and the discrimination accuracy can reach 100%.(3)Banana ripeness was distinguished based on characteristic variables.Using the continuous projection Algorithm(Successive Projections Algorithm,SPA)and Principal Component Analysis,Principal Component Analysis,PCA)two methods to extract the banana sample spectra data characteristic variables,the results show that the SPA method to the MSC and PA spectra data dimension reduction were obtained after pretreatment 9 and12 characteristic wavelength,The first five principal components(99.47% cumulative contribution rate)and the first four principal components(99.5% cumulative contribution rate)were obtained by PCA after pretreatment with MSC and PA respectively.In the discriminant model established by the feature wavelength extracted by SPA,the recognition effect of SVM model is lower than that of full-wavelength modeling with an accuracy of91%,while the recognition effect of LDA model is the same as that of full-wavelength modeling with an accuracy of 100%.In the discriminant model established by PCA extraction of principal components,the recognition effect of SVM and LDA model is the same as that of full-wavelength modeling,with an accuracy of 100%.