Abstract: This paper presents a method of multi-corn kernel embryos detection based on threshold segmentation and morphology. Corn kernel varieties identification is of great significance in the fields of agricultural production and crop breeding. In the seed market of China, the identification of corn varieties mainly depends on manual experience and measurement. In order to automatically, quickly, non-destructively identify kernel varieties, the study of automatic identification in a real time using machine vision technology is very active. Determination of the characteristics of the corn kernel is the first and the most important step of automatic identification. The corn kernel embryo is the most important part of the corn kernel. To analyze the characteristics of an embryo, an embryo must be separated from the corn kernel. The embryo detection speed and precision can influence the speed and precision of identification. In the paper, an algorithm based on threshold segmentation and morphology was proposed to segment embryos of multi-corn kernel at the same time, as a result of the deeper study of the identification. This algorithm was used to obtain multi-corn kernel embryos from a 2D digital image obtained by the scanner. It mainly included two parts, i.e. a maximum between-cluster deviation method (Otsu method) excluding pixels with zero value automatically, and improved open-close operation from morphology. Its process was as follow. In RGB color space, the multi-corn kernel embryos in the same image were segmented out at the same time by Otsu excluding pixels with zero value method based on the value of B(blue), in which the zero value pixels were auto-removed form histogram during processing. However, after segmentation, some corn kernel embryos showed a problem of lacking- segmentation or over-segmentation. To solve the problem, the improved open-close operation was used to repair the embryos. To validate the algorithm, four varieties of yellow corn which were common used in China were selected as study objects for our experiments. 45 samples were selected form each variety respectively. Then the total number of samples was 180. Every variety's digital image was obtained by scanner. Four images were obtained. They were processed respectively with the above-mentioned algorithm. The embryos from each different variety were detected. To validate the effectiveness of the detected embryos, two methods were used. First, area and perimeter of each embryo were measured respectively by machine computer and manual measurement. Linear regression analysis was done between the auto measured values and the manual values. The mean values of R2 were 0.9834 and 0.9578 respectively. Second, 6 shape-parameters which are perimeter, round degrees, ellipse strong and weak points axis ratio, rectangle degrees, and centrifugal rate were extracted from the embryo regions of 180 samples. Analyzing the data by K-means clustering method, the final clustering distances between different varieties reflected the difference in the visual of the embryos of the different varieties., and the checked out rate of the 4 varieties were 97.8%, 100%, 100%, and 100%. The efficiency of multi-corn kernel embryos detection was improved by 29.59% over single-corn kernel embryo detection. According to the experimental results, two conclusions were as follow: First, the auto-detected embryo region and the embryo region by manual experience and measurement were basically the same. The auto-detected embryo regions were effective. Second, the six parameters extracted from an embryo could be used to characterize the shape of the embryo. The results of this study provide favorable conditions for further study of the embryo characteristics of corn kernels, and provide a reference for the rapid and accurate identification of corn varieties.