Study On Quality Detection Of Prunus Mume Based On Hyperspectral And Machine Vision Technology

Prune has rich nutritional value and extremely high health benefits.In recent years,it has received widespread attention.The planting area is constantly expanding,and it has become the money bag for farmers in some areas to get rich.Due to defects in the growth process of prunes,fresh prunes are prone to fall injuries during picking,storage,transportation and sales,especially for fresh fruits with minor damage that are invisible to the naked eye,and are not easy to separate.If the slightly damaged prunes are not dealt with in time,it will accelerate the decay of itself and the intact prunes,which will directly affect the market value of the prunes.At present,in Xinjiang and other places that are mainly rich in prunes in my country,manual grading is still used for sorting,and the grading accuracy is unstable and the grading efficiency is also low.Therefore,identifying slightly damaged prunes and establishing a fast,simple and accurate quality grading system for fresh prunes have become urgent problems to be solved.In this thesis,the small French prunes in Taigu are the research object,and the quality of fresh prunes is graded by using hyperspectral technology to detect slightly damaged prunes and machine vision technology.The main research contents and conclusions are as follows:(1)In the range of 900-1700 nm in the full spectrum band,analyze and compare the effects of 5pretreatment methods(GF,MF,SG,Moving-Average and Baseline)on the overall effect of slightly damaged prunes and intact prunes at a distance of 2 meters and 3 meters from the land.The impact of discriminant accuracy.Five methods were used to preprocess the original spectral data of prunes,and the nonlinear extreme learning machine(ELM),linear partial least squares discriminant analysis(PLS-DA)and least squares support vector machine(LS-SVM)were established.For the discriminant model,the correct recognition rates for the three samples were 90.66%,88.44% and 91.56%,respectively.It can be concluded that the LS-SVM prediction and discriminant model established based on the original spectrum has achieved a better discriminant effect.(2)In the range of 900-1700 nm of the full spectrum,the RC,CARS and SPA algorithms are used to extract the characteristic wavelengths to establish a nonlinear extreme learning machine(ELM),a least squares support vector machine(LS-SVM)and a linear partial least squares respectively.Multiply(PLS-DA)discriminant model.The results show that: in the discriminant model of slightly damaged prunes and intact prunes at a distance of 2 meters and 3 meters from the land,the 13 characteristic wavelengths extracted by the SPA method,the established PLS-DA discriminant model has a good discriminant effect.The rate is94.67%.(3)Based on machine vision technology,the weighted average method is used to grayscale the prune image,and then the OTSU threshold segmentation method is used to segment the grayscale prune image,and the mask technology is used to separate the prune from the image background.The Canny operator is used to extract edge feature parameters from the prune image after background segmentation.The defect area of prunes was extracted by using color features.Two classifier models,post extreme learning machine(ELM)and least squares support vector machine(LS-SVM),are established through the extracted feature parameters and defect data.By predicting the test samples,the results show that the least squares support vector machine(LS-SVM)has the highest classification accuracy at 90.05%.(4)Design a quality grading system for fresh prunes.Using Matlab’s GUI toolbox,we realized the design of the user interface and developed and designed the quality grading system of fresh prunes.