Research On The Rapid Detection Method Of Jujube Moisture In Southern Xinjiang Based On Polarized Hyperspectral Imaging

The detection efficiency of traditional portable spectrometers cannot meet the growing needs of the Xinjiang jujube industry.And the hyperspectral sorter is limited by the size of the dark box,and cannot quickly detect the moisture of red dates.In order to improve the water detection efficiency of jujubes in southern Xinjiang,a series of hyperspectral polarization tests were carried out.The main conclusions of the research are summarized as follows: the two-way reflection characteristics of red dates,dark box data collection and modeling,the application of polarized images in the classification of red dates,and the rapid detection model of indoor red dates.(1)Unlike the dark box uniform light source,the reflectivity of the red date will be changed by the change of the incident zenith angle of the light source,the zenith angle detected by the detector and the relative azimuth angle when the date data is collected indoors and outdoors.The specular reflection parameters of the red dates do not reflect the internal moisture information of the red dates.Using the polarization two-way measurement system,the incident zenith angle(20°,30°,40°,50°,60°)and the detection zenith angle(30°,40°,50°,60°)And the relative azimuth(0°~ 230°)reflectivity.It can be seen that at a relative azimuth angle of 180°,when the detected zenith angle and the incident zenith angle are equal,its specular reflectance reaches the maximum value,and the hot zone of red dates is found.Because of the large error in the spectral modeling of the jujube hot zone,the subsequent experiments avoided the jujube hot zone.It provides guidance for the placement of light sources and detectors in subsequent experiments.(2)The spectrum collected by the dark box is stable,and is less affected by environmental factors.Based on the black box test,the follow-up will be extended to indoor and outdoor tests.The laboratory's hyperspectral sorter was used to collect dark spectral data of Nanjiang jujube,and then the black and white correction was performed.The moisture content of red dates was determined by drying and mass reduction method.A variety of spectral preprocessing methods are used to preprocess the original spectrum of red jujube,and the continuous projection algorithm(SPA)and the competitive adaptive weighting algorithm(CARS)are used to optimize the wavelength,respectively.Finally,compare partial least squares(PLS)and BP neural network modeling.After many adjustments,the model established by the BP neural network was found to be optimal.The correlation coefficient(R)was 0.9321,the corrected standard deviation(RMSEC)was 0.5841,and the predicted standard deviation(RMSEP)was 1.0281.But because the BP neural network is very sensitive to the initial weight,it is easy to converge to the local minimum.Therefore,a more stable partial least square method is used as the modeling method for the indoor rapid detection model.(3)Data collection was carried out outdoors using near infrared polarized hyperspectral sorter and visible light polarized camera.Respectively calculate the data to obtain S0,S1,S2,Dolp and Orient images,and perform RGB synthesis.Support vector machine(SVM)was used to classify the enhanced image,and the jujube was successfully separated from the complex background.The total accuracy is 0.9585 and 0.9932,respectively.It provides technical support for mask and position spectrum data extraction for rapid detection of jujube moisture.(4)According to the bidirectional reflection characteristics of red dates,set the relative azimuth angle of the camera and the halogen lamp at 40°,the detection zenith angle of the camera at 30°,and the light source incident zenith angle at 35°,avoiding the hot zone of red date reflectivity.Place the red dates on the indoor iron shelf to collect the data.The SVM is used to classify the polarized image.Extract location information,spectrum information and build a mask of red dates.Finally,the model is established with acorrected standard deviation(RMSEC)of 1.0683,a predicted standard deviation(RMSEP)of 1.0805,a relative analysis error(RPD)of 2.3345,and a correlation coefficient(R)of 0.9156,which meets the accuracy of moisture detection compared with the modeling accuracy of the dark box Claim.Apply the established optimal model to each pixel of red dates in the data to get the gray map of red dates.At this time,the gray value of the red jujube image corresponds to its moisture value.Then apply false color pseudo coding technology,use different colors to represent different moisture values on the red dates,and express the quality distribution characteristics of the red dates.Realize the visualization and rapid detection of red jujube moisture.