Research Of Detection About Fresh Jujube Pests Based On Near-Infrared Spectroscopy And X-Ray Image

Jujube has high nutritional value, but the pests is very easy to happened in planting. Many factors,such as the bug is smaller、there are splash and pest area is uncertain, lead to the internal quality is difficult to determine.Pest detection of jujube using of methods of artificial identification, not only costs a lot of manpower, but also is difficult to find. With taigu normal (150) and pests (150) of huping jujube as the research object, while the average normal of huping jujube was divided into three groups, the number of each group is50, in order to meet the need of sampling spectral information in the part of calyx、terrier、central; pests pot of jujube, which bug is located in the middle of calyx、terrier、central, have50respectively. Gather near infrared spectrum and X-ray image information about fresh jujube.According to the proportion of4:1, the sample were randomly divided into the training sets and the predictive sets.Chemometrics method was used to preprocess and set up classification discriminant model about the information of near infrared spectral; processing X-ray image use the image denoising. image enhancement image segmentation and morphology processing method successively to extract the insects features of image, then set up classification discriminant model,reach the purpose of determining the optimal model of near infrared spectrum detection. establishing discriminant model of X-ray image and realize the discriminant of fresh jujube pests. The main research contents and results are as follows:(1) In data preprocessing of near infrared spectrum analysis about fresh jujube pests, use12kinds of methods, respectively,3point smoothing、5point smoothing、7point smoothing、9point smoothing、MSC、SNV、SG smoothing、Baseline. first derivative、second derivative. MSC+first derivative and SNV+first derivative. Through PLS modeling analysis, we found pretreatment effect of SNV is best, Re、Rp、RMSEC、RMCEP of getting about the training sets and the predictive sets are respectively0.908641.0.874301.0.208789.0.243787.(2) In data modeling of near infrared spectrum analysis about fresh jujube pests, useSNV-PLS、SNV-PCR、SNV-LS-SVM、SNV-PCA-PLS、SNV-PCA-PCR、SN V-PCA-LS-SVM、SNV-PCA-BPANN、SNV-SPA-PLS、SNV-SPA-PCR、SNV-SPA-LS-S VM to classify between Normal jujube and insects jujube. The study found that modeling effect of SNV-SPA-LS-SVM is best, the accuracy of discriminant is100%.This provides the basis for study of fresh jujube pest on-line detection.(3) In denoising of X-ray images analysis about fresh jujube pests, using Matlab2008a platform, use3×3median filter、5×5median filter、7×7median filter、3×3average filter、5×5average filter、7×7average filter、3×3wiener filter、5×5wiener filter、7×7wiener filter、3×3gaussian filter、7×7gaussian filter、11×11gaussian filter、butterworth low-pass filter、global threshold wavelet denoising、gobal threshold wavelet packet denoising to pretreatment, there are15kinds of denoising methods. The study found that the MSE and MAE of3x3gaussian filter is least, they are respectively1.8352、1.0048; PSNR is biggest, it is42.8924, this method is best.(4) In extracting pest feature of X-ray images analysis about fresh jujube pests, grey of image after filtering value has been extended to [0255] range for image enhancement using the method of gray scale linear transformation.through satistical analysis of fresh jujube and pest area size, this research determines automatic recognition template of suitable for the segmentation is [L/2B/2] and the center is the center coordinates of the enhanced image. We get pests center area with this template, use iterative arithmetic can get the best segmentation threshold of each fresh jujubes, determined that the best segmentation threshold is61.5through statistical analysis, get the threshold image of pests center area.Otsu’s method was used to solve adaptive threshold in ord tonegative image of jujube’s pest center area, then the subtraction of Threshold image and negative image was used as a preliminary image segmentation. Based on these, we used6×6corrosion、3×3expansion、logical and operation between image of expansion and image of preliminary segmentation、cavity filling to get pests feature area of huping jujube. Build the evaluation index of pixel ratio, the image of(5) In extracting pest feature of morphology disposal, in discriminant classification between normal and pest of jujube, we use the pixel ratios to set threshold.Through the statistics of the pixel ratio about the training sets (240), we use overall discriminant accuracy up to the highest as standard to determine the optimal threshold that is3.75%. In this case, the overall discriminant accuracy of the training sets is87.1%.We use this model to predict60samples about the predictive sets, the overall discriminant accuracy of prediction is87.1%.