Monitoring of planted lines for breeding corn using UAV remote sensing image

Abstract: The planted line is one of the important phenotypic parameters for breeding corn. And extracting the planted lines of breeding corn quickly and nondestructively is vital to corn breeding trials and research. Unmanned aerial vehicle (UAV) remote sensing technique has advantage in extracting phenotypic parameters of corn canopy, which can be used to extract the phenotyping parameters in large area for breeding corn. This research is aiming at extracting the planted lines for breeding corn using UAV image acquired from DJS1000+ UAV platform by calclulating the super-green feature and Hough transform of the UAV images of 3 different growing seasons, i. e. seeding stage, jointing stage, and maturation period. The first step is calculating the super-green features of 3 UAV images, and then binary optimization and morphological open operation are performed so as to separate corn canopy and soil background. In addition, we use 3 different window sizes, i.e. 1?15, 1?25, 1?50, to search and detect the locating middle points for extracting planted line of breeding corn in 3 different growing seasons. Next step, the center point of corn ridge line is identified by projecting the segmented objects. The judging conditions of left border and right border for center points extraction are as follows: The points will be classified as left border points if the pixel mean is greater than pixel of column j-1 and the pixel mean is less than pixel of column j+1; and the points will be classified as right border points if value the pixel mean is less than pixel of column j-1 and the pixel mean is greater than pixel of column j+1; the central location will be labeled as the central point of planted line. Finally, the line of breeding corn is identified by Hough transform of the above center points of corn ridge line. Hough transform is casting all the points in image domain to the points in transforming domain, and one point in image domain is corresponding to one sine curve in transforming domain. Therefore, the sine curves that are corresponding to the points located on the same line will be intersected at one same point, which will establish the line in image domain. We select 3 breeding corn plots in 3 different growing periods i.e. seeding stage, jointing stage, and maturation period. The results indicate that the planted line of breeding corn in jointing stage can be extracted more accurately than that in seeding stage and maturation period. There were 74 lines used in planted line extraction experiment in jointing stage, and our extraction results are 74, 74, and 105 lines using 3 different window sizes respectively. In this growing stage, the corn plants are big enough for planted line extraction and the leaves between different lines do not connect meanwhile. There were 43 lines used in planted line extraction experiment in seeding stage, and our extraction results are 42, 45, and 58 lines using 3 different window sizes respectively. The leaves between different lines have already connected, which affects the planted line extraction. And there were 44 lines used in planted line extraction experiment in maturation stage, and our extraction results are 46, 40, and 49 lines using 3 different window sizes respectively. In this seeding stage, the leaves are too small (1-2 cm) and the planted line extraction is more difficult. So our method can be used to extract the number of breeding corn lines, and the optimal growing stage for planted line extraction is jointing stage and the window size for searching and detecting the locating middle points should be similar to the row spacing of planted corn.