任意姿态玉米种子定位方法研究

Localization of Maize Seed at Random Position

为准确提取玉米种子不同区域色彩特征和形状特征,解决机器视觉检测中需以特定装置固定种子位置以及人工规则摆放的问题,提出一种任意姿态玉米种子定位的方法。该方法首先对采集的任意姿态的玉米种子彩色图像进行预处理,实现对单粒玉米种子图像的分割提取;然后利用B通道阈值对单粒种子图像进行分割,计算黄色部分像素面积比例,判断出种子无胚芽侧面并对其归类;其次,对种子轮廓图像坐标以遍历方式计算轮廓间最长距离的方法获取种子尖端点坐标,并通过一阶中心矩提取质心坐标,得到玉米种子图像倾斜度,求得图像旋转角度θ°;最后,对玉米种子图像进行θ°仿射变换(Affine transformation)旋转,获取玉米种子水平放置图像,为进一步提取色彩特征和形状特征提供准确定位。试验表明,该方法可将任意姿态的玉米种子进行有效旋转定位,具有较好的实用性,同时该方法对于其他需要定位的种子图像同样适用。

In order to extract color features of different regions of maize seed and solve the problem that seeds need to be put by manual regular operation and specific device in machine vision detection,a method of maize seed location at random position was put forward. Firstly, the color image of maize seeds at random position was preprocessed and single seed images were extracted. Then single seed was segmented by B component threshold to calculate area of yellow part and the ratio of endosperm area in all seed, judging the classes of maize side. Then, the longest distance L between all contour-coordinates could be calculated by traversal behaviors that were used to obtain tip coordinate of grain. The centroid of grain image counted by first order central moment of grain contour combines with tip coordinate to calculate gradient of maize seed image and rotation angle θ° Finally, the grain image was rotated in certain angle by affine transformation to get horizontal position image of maize seed which could provide accurate location of grain seed for extracting the color features and shape characteristic further. The result showed that this algorithm could rotate the maize seed at random position to horizontal position effectively, and that it was suitable for other grain seed which needs to locate in detection and recognition.