摘要
为解决玉米育种激光切片取样机器人设计中玉米种子自动定向问题,该文提出了一种计算机视觉与姿态调整装置相结合的定向方法,并使用此方法做了验证试验。使用MATLAB神经网络模式识别工具箱,以种子的部分特征参数作为神经网络的输入,以预定义的种子形态和姿态为输出,建立模型并对模型进行训练,训练完成的模型用于种子姿态识别,姿态调整装置根据识别结果将种子调整到理想姿态。对497组样本进行训练,得到形态识别准确率97.8%,姿态识别准确率99.8%,识别速度1.3 s/幅,能满足设备的设计要求,验证了姿态识别方法的准确性和姿态调整装置设计的合理性。该方法可为玉米种子自动定向问题提供解决方案,为自动育苗机中自动定向装置的设计提供参考。
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.
出处
《农业工程学报》
EI
CAS
CSCD
北大核心
2016年第4期40-47,共8页
Transactions of the Chinese Society of Agricultural Engineering
基金
辽宁省自然科学基金项目(2014020083)
辽宁省农业攻关及成果产业化项目(201404236)
关键词
种子
神经网络
模型
玉米育种
种子自动定向
机器视觉
姿态识别
seed
neural networks
models
maize breeding
seed auto orientation
machine vision
posture recognition