Study On The Method Of Identification And Classification Of Jujube In South Of Xinjiang Image Based On The Hyperspectral Technology

Jujube,as an edible crop in Xinjiang,is also an important cash crop.It is loved due to its high nutrition and medicinal value by the majority of consumer groups.Detection of external quality of the jujube is an essential link before production and processing,So for the production and processing of the jujube quality,external quality testing of the jujube is very important,For the study of external quality testing and fine of the jujube in Xinjiang.In this paper,The characteristic bands of reflected absorption peaks are used to investigate the conditions of indoor acquisition at different angles and outdoor conditions at different angles.hyperspectral image information is used to identify diseased jujube,Includes branches and jujube leaf of jujube(not falling and falling of jujube),and the detail contents are follows:(1)Firstly,analysing the four normal indoor jujube reflection characteristics absorption band,and to applicate 989 nm and 1443 nm band ratio to achieved multi target jujube image recognition spectral.According to the reflection of four obvious reflection absorption peak indoor jujube by using the average difference analysis in order to reduce indoor collection of stray light on different angles reflectance collected as the incident light interferencence(30 degrees,45 degrees,60 degrees),and obtained for 60 degrees(Included angle with horizontal plane)light reflection maximum average value is 0.074638,the information is the largest and more suitable for the description of different angle.It is obvious that the angle of incident light reflection spectra are different.Last using the hyperspectral image technology to acquire the information of lesion of jujubeby before black and white correction.,and there are three kinds of image enhancement methods: Principal Component Analysis(PCA),Characteristic of brand ratio and HSV.Then,to classify and evaluation of the lesion areas of jujube based on Support vector machine.Finally,according to classify and analysis of the result of lesion areas of jujube,The total classification accuracy is 99.115%,Kappa coefficient is 0.9879.The main components of 120 diseased jujube were analyzed and classified,and the pathological regions of diseased jujube could be identified and classified.The average total accuracy was 97.2%,and the Kappa coefficient was 0.96.it shows the Principal component analysis(PCA)can make the best effect.It can realize Multi-target identification and classification of diseased jujube by Hyperspectral Image Technology and the area of lesion can be calculated.(2)In order to reduce interference of outdoor collection of stray light,using the laboratory design of measuring instrument of two polarization to simulate outdoor conditions on different angles(0 degrees,10 degrees,20 degrees,30 degrees)of reflection information acquisition of the average value reached 0 degrees(perpendicular to the angle of reflection red dates)information the most,is the most suitable acquisition point of view,illustrate the reflection information collected from different angles,the reflection information acquisition is different.Different angles in outdoor applications of hyperspectral camera Jujube(0 degrees,10 degrees,20 degrees,30degrees)of the reflectance spectra of four Jujube,according to the obvious reflection absorption peak of 939 nm,1130nm,1368 nm,1718nm was obtained by analyzing the reflection difference of0 degrees(the camera level parallel to the ground)the average reflection maximum difference reached 0.3715,compared to the indoor simulation experiment to optimize the best outdoor multiangle acquisition point of view,that is more appropriate in the outdoor 0 degree angle.The acquisition of hyperspectral image application in jujube obvious absorption the peak of 1130 nm,the characteristics of 1368 nm band ratio image,construct the RGB pseudo color image of single band image NDVI image and 1718 nm,Jujube will be separated from the background Jujube feature information is highlighted,select the outdoor spectrum to jujube The establishment of outdoor spectrum database.Hyperspectral image information of jujube is acquisited by outdoor equipment of hyperspectral.Constructing pseudo color after the black and white correction.It can determine the category of image classification The classification category is jujube,stems,land,background.Four spectral databases of jujube are established,they are jujube,leaves,stems and land.Hyperspectral image of jujube will be utilized by Normalized Difference Vegetation index,Independent Component Analysis,Minimum Noise Fraction Rotation,Principal Component Analysis.By comparing the spectral image of true observation during collection.Principal Component Analys is the best effect and also be the most suitable processing method.The classified images will take advantage of support vector machine,Back propagation artificial neural network,Maximum likelihood,Minimum distance,Mahalanobis distance,Parallelepipedon,Spectral mapping of the Supervised classification.Finally,according to the analysis and evaluation,Finally,the total classification accuracy is 99.668%,and Kappa coefficient is 0.9937.it shows the effect of Support vector machine based on Principal component analysis is better than other methods and it is suitable for processing the identification and classification of hyperspectral image of jujube garden.