Computation method of phenotypic parameters based on distribution map of kernels for corn ears

Phenotypic traits of corn ears are important parameters for maize breeding and production forecast. A phenotypic computation method based on distribution map of kernels for corn ears is presented to comprehensively analyze the geometrical, quantitative, color and texture traits of corn ear and its kernels. A phenotypic detection system of corn ears, which consists of stepping motor, charge coupled device (CCD) camera, light-emitting diode (LED) back lights, image acquisition card and semi-closed box, is designed to capture main side images of corn ear. Corn ear is fixed vertically on a turn table driven by stepping motor, and thus image sequence of corn ear can be captured from designated angles and covers the entire surface of corn ear. In this study, 4 orthotropic images are captured to build three-dimensional reference frame of corn ear. Firstly, axial and radial distortion corrections are successively applied to image sequences and generate standard image sequence of corn ear. Therein, axial distortion correction regularizes the heights of corn ear for image sequence to their average height, and radial distortion correction is used to recover size and shape of kernels on the surface of corn ear, since those kernels lie on border regions of corn ears which have obvious distortion. A dedicated image segmentation method, which has utilized geometrical and color traits of kernels of corn ear to the greatest extent, is then used to extract all kernels of corn ear from corrected image sequence. Meanwhile, contour lines and split lines of corn ear in image sequence are calculated based on pixel scale, and used to generate the mapping relationship among the images of corn ear. Contour lines of corn ear can be used to generate a three-dimensional surface model by transforming the coordinates of lines from two-dimensional to three-dimensional space, and then the three-dimensional model of corn ear is used to calculate geometrical traits, e.g. perimeter, surface and volume. Moreover, contour lines of corn ear can also output corresponding split lines which split each corn image into different regions, and further classify kernels into different types according to position relationship between kernels and split lines. On the basis of kernel scale, the classified kernels from segmented image sequence can be assembled together to generate a distribution map which describes entire surface kernels of corn ear. According to the distribution map of kernels and segmented image sequence, the quantitative and geometrical traits of kernels, such as rows per ear, kernels per row, total kernels and kernel thickness can be accurately calculated using Delaunay and Bellman-Ford methods. The proposed method and system can simultaneously detect multiple types of phenotypic traits from image sequence of corn ear, and have higher accuracy in almost all phenotypic traits than the detection method based on single profile image of corn ear. Experimental results demonstrate that the computed traits have good consistence with the observed values, and the average computation accuracy of main traits, i.e. rows per ear, kernels per row, total kernels, ear length and ear diameter, can respectively reach 98.231%, 94.351%, 96.921%, 98.956% and 98.165%. Thus, the proposed method can be applied for precise phenotypic detection and breeding of corn ears.