Research On Disease And Pest Detection Method Of Blueberry Based On Hyperspectral Imaging

In recent years,the blueberry industry has developed rapidly in China,and blueberry fruit has rich nutritional value and is deeply loved by consumers.Blueberries are not easy to store after harvesting.During the period,slightly rotten blueberries and pest-infested blueberries are prone to decay,causing secondary losses.Therefore,the sorting of blueberries after harvesting is very important.At present,nondestructive testing of fruit pests and diseases is a hot research topic,including machine vision technology,spectral analysis technology and hyperspectral imaging technology.Machine vision technology is greatly affected by the surface color of the tested sample.Spectral technology needs to accurately determine the area of interest in order to extract accurate spectral information for analysis and detection.Hyperspectral imaging technology combines image processing and spectral analysis technology to make up for the shortcomings of machine vision technology and spectral analysis technology.Many domestic and foreign researchers use hyperspectral imaging technology to detect fruit pests and diseases,and have obtained better detection results.Therefore,this paper takes blueberries from a blueberry planting base in Shenyang city as the research object,and carries on the non-destructive testing of blueberries and fruit fly pest blueberries based on hyperspectral imaging.The research results provide important reference for the development of the online non-destructive testing system for blueberries.(1)In this paper,spectral information segmentation(SIS)is proposed to segment slightly rotten blueberry and insect pest areas,and extract spectral information.The original spectral data of blueberry were processed by SG Convolution smoothing method(SG),standard normal variable variate(SNV)and multiplicative scatter correction(MSC).Radial Basis Function(RBF)neural network model was established to evaluate the detection effect of three kinds of pretreated spectral data and original spectral data.The results showed that the spectral data pretreated by Scattering Correction(MSC)had the best detection effect and could be used for the following detection of blueberry diseases and insect pests.(2)The regional feature selection(RFS)method is proposed in this paper.The full-band spectra are divided into the first region of visible light(500nm-760 nm)and the second region of near infrared(760nm-1000 nm).Then the characteristic wavelengths of blueberries are extracted by SPA and CARS-IRIV.The SPA method extracts three characteristic wavelengths in the first region and three characteristic wavelengths in the second region.CARS-IRIV method extracts three characteristic wavelengths in the first region and four characteristic wavelengths in the second region.Based on these characteristics,LDA,RVM and RBF models were established to detect blueberry rot.By comparing the detection results,the CARS-IRIV-RBF model of the combination of the first and second regions was chosen as the best detection method.The detection rate of blueberry diseases in training set was 90%,and that of blueberry diseases in test set was 90%.(3)By comparing single-band images,images of blueberry pests in 890nm band were selected to acquire texture features of blueberry stem,calyx,pest and normal area.LDA,RVM and RBF models were established as input vectors to detect blueberry pests.The results show that when the energy,entropy,moment of inertia and correlation in the direction of 0 degree are used as input vectors of the model,the accuracy of RBF detection can reach more than 85%.CARS-IRIV and SPA methods were used to extract characteristic wavelengths of blueberry pests.Five characteristic wavelengths were extracted by CARS-IRIV method and nine characteristic wavelengths were extracted by SPA method.Then LDA,RVM and RBF models were established to detect insect pests in blueberry.The results show that the RBF model based on CARS-IRIV extracting the reflectivity corresponding to the characteristic wavelength has the best detection effect.The RBF model was established to detect insect blueberries by combining the texture features in 0 degree direction with the reflectivity corresponding to the characteristic wavelength extracted by CARS-IRIV.The results showed that the detection rate of insect blueberries was higher,and the accuracy rate was over 88%.