Study On Apple Quality Detection System Based On Spectral Analysis Technology

Apple,known as the king of fruits,contains some polyphenol compounds and essential trace elements that can effectively remove metabolites in the body,reduce blood sugar,blood lipids,etc.,and is the natural killer of cardiovascular and cerebrovascular diseases.But there are many kinds of apples on the market and the quality is far from the same.The important indicators for evaluating apple quality include damage and sugar content,which are directly related to the taste of apple.In order to ensure the economic benefits of producers and middlemen,and also meet the needs of consumers,it is urgent to improve the quality detection system to provide reference value to all demanders,but the cost of non-destructive testing instruments on the market is expensive,Therefore,it is of great significance to have effective detection accuracy on the basis of reducing the cost of apple quality detection system.This research takes Red Fuji apple as the research object,and the main research contents include the research on the method of non-contact detection of apple invisible damage based on hyperspectral imaging system;Research on nondestructive detection of apple sugar content based on near-infrared spectroscopy;Based on the research of damage detection methods,the multi-spectral detection system that can be used in industrial production lines is studied;Based on the method of nondestructive detection of apple sugar content,a multi-wavelength portable apple sugar content meter was designed.(1)Firstly,the hyperspectral full-band image is analyzed by principal component analysis(PCA)and minimum noise separation(MNF)to find the most obvious principal component of the damage area,and then the feature wavelength is filtered based on competitive adaptive reweighting(CARS)algorithm and continuous projection(SPA)algorithm respectively,and the damage degree is classified by combining support vector machine(SVM).The experimental results show that the classification accuracy reaches86.67%and 91.67%respectively,Finally,the image corresponding to the characteristic wavelength is combined into a multi-wavelength image,and the MNF transformation is performed.The result image is similar to the full-band transformation,and the overall recognition rate is 98.125%.It is proved that the image based on the characteristic wavelength can be used to replace the full-band image for the detection of early apple invisible damage.(2)Based on the sugar analysis of the near-infrared spectrum,the best pretreatment method is selected first.The research results of the selection of the pretreatment algorithm show that the wavelet de-noising-multiple scattering correction-partial least squares(DWT-MSC-PLS)model shows better predictability.The number of determination factors(R~2)of the training set and the test set are 0.8713 and 0.8511,respectively,and the root mean square error(RMSEC)of the training set and the root mean square error(RMSEP)of the test set are 0.9413 and 1.1915,respectively.Therefore,DWT-MSC preprocessing method is adopted for subsequent modeling.On this basis,CARS and mutual information(MI)algorithms are used to screen the characteristic wavelengths,and then PLS model is established and compared with the full-band PLS model.It is found that the filtered characteristic wavelengths can effectively improve the accuracy of the model and reduce the amount of data.The characteristic wavelength model of mutual information screening has higher accuracy,and the selected wavelength accounts for 47.85%of the total wavelength.Compared with CARS-PLS,the number of characteristic wavelengths of MI-PLS has increased by 64.55%,but R~2 has increased by 0.0472,Provide theoretical support for subsequent hardware design.(3)A multi-spectral damage detection system based on hyperspectral imaging technology,and a portable apple sugar meter based on multi-wavelength.Based on the research of hyperspectral images,a multispectral analysis system is established,including the design of imaging system,the establishment of classification model and software development,and the algorithm is transplanted to FPGA for image recognition to accelerate.The results show that the accuracy rate of damage identification on the feature image is 97.5%.Compared with GPU,the operation speed is increased by 33.552%after transplantation to FPGA,and the accuracy rate is slightly reduced to 93.75%.Finally,the three damage levels are classified by SVM,and the accuracy rate is 88.33%,which proves the effectiveness of the equipment.(4)When designing the multi-wavelength portable apple sugar meter based on the characteristic wavelength of apple sugar screened by the near-infrared spectrum,it mainly involves the hardware design,appearance design and multiple linear regression function analysis.On the basis of reducing costs,it is compared and analyzed with similar devices with better effects on the market.The results showed that the R~2 of multiple linear regression analysis reached 0.9388,the root mean square error was 0.2633,and the model effect was good.Compared with ATAGO nondestructive saccharometer,the accuracy difference between the two was 3.28%,which reached the expected effect.