Abstract: Abstract: Molecular marker breeding is often used in modern maize breeding. Breeding section is an analytical sample used in modern molecular marker breeding, and maize breeding section is taken from the endosperm tissue of the endosperm end of the corn seed, which is about 15 mg and can be cut by hand and also by machine automatically. The corn breeding laser chip sampling robot is a device for automatic production of maize breeding section. In the process of making section, the accuracy of the automatic orientation determines the location and size of the section, and is a key link in the whole process. In order to solve the problem of automatic orientation of maize seeds in the design of corn breeding laser chip sampling robot, one method of corn seed orientation combined with computer vision and posture adjustment device is proposed in this paper, and the validation test of this method is done. In short, the method is to obtain the current posture of the seed from the seed image information at first, and then use the posture adjustment device to adjust the seed from the current posture to the ideal posture. The main content of the paper is the identification method of the seed posture and the design of the posture adjusting device. Based on the shape and posture of maize seed, through analyzing and comparing the differences of main characteristic parameters abstracted from binary image of those seeds under different shape and position, the obvious characteristic parameters are selected as the effective characteristic parameters, and the effective form and posture parameters include the ratio of perimeter to area, the roundness, the ratio of the minor axis to the major axis, and the proportion of area size of the sub regions, which are divided into 9 rectangular areas (3×3) uniformly from the target binary image. Using these effective characteristic parameters as input parameters of neural network, and the type of the predefined target shape and posture as the output parameter, the neural network model is established and trained using the MATLAB neural network pattern recognition toolbox. Finally, the training completed neural network model will be used as a classification model for the identification of seed morphology and pose. According to the output of the classification model, the angular deviation between the current seed pose and the ideal posture can be calculated, and the posture adjustment platform drives the rotation of the seed according to the deviation value, which makes the seed's posture close to the ideal posture. A clamping centering device is designed in posture adjustment platform, and to some extent, it can make up for the deviation between the finiteness of neural network recognition and the randomness of seed posture. The following is the result of the seed orientation test: the 497 groups of samples are trained; the morphology recognition rate is 97.8%, the posture recognition rate is 99.8%, and the recognition rate is 1.3 s per image, which can meet the design requirements of the corn breeding laser chip sampling robot, and verifies the accuracy of the method and the rationality of the design of the posture adjusting device. The research can provide a solution for the automatic orientation of the seeds in the design of the corn breeding laser chip sampling robot, and also a reference for the design of automatic orientation device, such as sunflower seed and gourd in seeding machine.