Monitoring Of Phenotypic Characteristics Of Millet Based On Image Processing

Millet is the main crop of dry farming in China.It originated in China.As one of the small grains,it has high nutritional value.Plant phenotype refers to a series of physiological and biochemical traits shown by crops during their growth and development,which can reflect the current crop growth status to a certain extent.Therefore,rapid measurement of plant phenotype plays a significant role in improving crop yield and quality.However,at present,the phenotypic measurement of millet in China is mainly based on manual measurement,which requires a lot of time and labor,and human eye judgment is also subjective,which makes the phenotypic measurement inefficient and poor accuracy.The continuous integration of image processing technology and agriculture can improve the problems existing in the above traditional manual measurement,and realize the intelligent,automatic and efficient measurement of millet phenotype.To jin valley 21 of this study was to test materials,phenotypic measure mainly divided into two parts,field and indoor,by digital camera respectively for millet DTW group and millet indoor plant RGB image,then USES the gray,binarization and morphological operation and skeleton extraction such as common image processing technology,realization of foxtail millet field canopy of the image segmentation and stem leaf of foxtail millet indoor image segmentation,and through the segmentation result,study the excellent algorithm to calculate millet leaf area index,biomass,leaf length,leaf width,plant height phenotypic parameters,to achieve rapid measurement in the phenotype of foxtail millet.The main research results are as follows:(1)Canopy extraction of field millet.Three algorithms,namely super green segmentation,Kmeans clustering segmentation based on Lab color space and Kmeans clustering segmentation based on H component,were adopted to extract canopy from four kinds of millet canopy images: cloudy days,complex background with shadows,uneven illumination and reflective dew and rain water.The experimental results show that the three algorithms can extract the millet canopy images on cloudy days and with complex shadowy background completely,and the segmentation accuracy reaches more than 93%.For images with uneven illumination,ultra-green segmentation has the worst effect,while K-means clustering based on Lab space and HSI space has relatively better segmentation effects,which are 93% and 96%,respectively.For the images reflected by dew and rain,the K-means clustering based on H component has the highest segmentation accuracy,reaching 97%.(2)Stem and leaf segmentation of indoor millet.After obtaining the original indoor millet image,it is necessary to pre-process the original RGB image to achieve binarization.Then,the skeleton extraction and the skeleton thining algorithm based on Zhang are adopted to obtain the final skeleton model of millet,seek the intersection points of the skeleton,and finish the stem and leaf segmentation and the segmentation of each leaf through leaf marking and intersection points.The results showed that the algorithm could realize the stem and leaf segmentation of millet quickly and accurately,and the segmentation effect of millet at different growth stages was good.(3)Phenotypic measurement of Datiantructus.In full,on the basis of extracting corn canopy image are calculated separately,and the canopy coverage CC standardized value b,green and blue index EXG standardized value g,green,red index EXR,a green minus ultra red index EXGR,brightness INT,standardized value r seven visible light red color index,the above image characteristic and the corn leaf area index and biomass of correlation analysis,selection of high correlation characteristics,establish millet leaf area index and the biomass of SVM,PLSR,BPNN and RF four kinds of prediction model.The results showed that,except for INT and EXR,the other image features were significantly correlated with leaf area index and biomass of foxtail millet,among which CC,B,EXG and EXGR were positively correlated,while R and G were negatively correlated.The BPNN model constructed in the estimation model of leaf area index has the best prediction effect,and the R~2 and RMSE of this model are 0.8529 and 0.3121 respectively.The biomass estimation model was similar to the LAI estimation model,and the BPNN model had the best prediction effect,with R~2 of 0.7924 and RMSE of 0.0891.As the number of features decreases,the accuracy of the model decreases gradually.(4)Laboratory phenotype measurement of millet.Leaf length,leaf width and plant height were the main parameters for indoor foxtail millet phenotype measurement.Each phenotypic parameter measurement algorithm was studied respectively.The leaf length was mainly represented by the leaf skeleton.Blade width is to find the Angle value of blade width by making the vertical line in the direction of blade length,and find the position of the center of gravity of blade.Draw a straight line with the known Angle and center of gravity,and the line intersects the blade edge to form two key points.The distance between the two key points can represent the maximum width of blade.The plant height of millet needs to find the minimum enclosing rectangle of the plant and calculate the length of the minimum enclosing rectangle,which can represent the plant height of the current plant.The experimental results showed that the leaf length,leaf width and plant height measured by the above algorithm were all accurate,and the relative errors were3.81%,6.05% and 3.72%,respectively.