Research On Ripeness Detection Method Of Blueberry Based On Hyperspectral Imaging

Because of its unique flavor,rich nutritional value,blueberries are deeply loved by consumers in China.However,in actual production,unripe fruits and overmature fruits will be picked together.In order to sell blueberries in the future and improve the quality of blueberry processed products,blueberries need to be sorted.Therefore,this study is of great significance to the on-line detection and classification of blueberries by studying the maturity detection methods of blueberries.In this paper,the fruits from the blueberry planting base in the southern district of Shenyang City as the research object.It is based on hyperspectral imaging technology,and the blueberry ripeness were classification discrimination from spectral information and image information in order to achieve a fast and non-destructive blueberry ripeness detection,and build blueberry ripeness online detection provide a theoretical basis.The main research contents and results are as follows:(1)The modeling results of different preprocessing methods are compared.Multiplicative Scatter correction(MSC),Standard Normalized Variate(SNV)and Savitzky-Golay Smothing(SGS)were used to preprocess the original spectral,and then the BP neural network model was established.The results show that the BP model based on SGS preprocessed spectral data has the best prediction performance.(2)The ripeness of fresh blueberry was classified and distinguished based on full-wavelength information.Spectral information preprocessed by SGS was used.The PLS,BP and SVM discrimination models had been built respectively on the basis of the data of full-wavelength information of different ripeness(unripe,mature and overmature).The results show that the BP model based on full-wavelength can get the best discriminant result,the identifying rate of unripe,mature and overmature was 96.67%,93.34% and 89.34%,respectively.(3)The ripeness of fresh blueberry was classified and distinguished based on characteristic wavelengths information.After collecting the hyperspectral image of blueberry,the blueberry's soluble solids content and firmness were measured.The characteristic wavelength of soluble solids content,firmness and the combination of soluble solids content and firmness were extracted using SPA,SMLR and CARS,respectively.Then establish the predict model of the characteristic wavelengths of soluble solids content,the characteristic wavelengths of firmness and the characteristic wavelengths of the combination of soluble solids content and firmness using BP algorithm.Finally,the PLS,BP and SVM discriminant model for the ripeness of blueberries were established based on soluble solids content,firmness and the combination of soluble solids content and firmness.The results showed that the BP discriminant model of blueberry ripeness established by multi-stage SPA-SPA was the best prediction model,the identifying rate of unripe,mature and overmature was 98%,92.65% and 92%,respectively.(4)The ripeness of fresh blueberry was classified and distinguished based on the fusion technology of multi-feature information.After the blueberry image was collected,the image was subjected to median filtering and denoising.Then the RGB,HSV and Lab models were used to extract color features,the PLS,BP and SVM discriminant models were eatablished for RGB,HSV,Lab and the fusion of spectral and image.The results showed that the BP classification discriminant model established using SPA-SPA+RGB+HSV as model input was the best blueberry ripeness discrimination model,the identifying rate of unripe,mature and overmature was 96%,96% and 92%,respectively.Through the analysis of the above experimental results,the BP discriminant model of the ripeness of blueberry established by using SPA-SPA + RGB+ HSV as the input vector has the best performance among all the discriminant models,and the discriminant accuracy of all kinds of samples is the highest,which provide a theoretical basis for that follow-up construction of blueberry online detection.