摘要
为解决脱穗之后的玉米种粒三维特征的获取和分析这个重要而又困难的问题,该研究基于机器视觉技术开发了一种融合三维特征的玉米种粒考种装置。装置通过2块相互垂直的标定板来保留倾斜影像中的空间信息和标定数据,并据此计算三维数据。基于装置的标定数据,用图像处理的方法获取玉米种粒的长度、宽度和厚度数值。通过透视变换,从倾斜摄影图像分别得到水平正摄和垂直正摄的图像;玉米种粒轮廓的长轴和短轴以旋转盘的直径为参考进行计算;玉米种粒的厚度以垂直方向棋盘格标定数据为参考进行计算。按照10帧/s的帧率和1 280×720的图像分辨率启动图像记录系统。选取180粒不同品种的玉米种粒进行试验。种粒长轴、短轴和厚度测量的均方根误差分别为1.86、1.28和0.741 mm;决定系数分别是0.849 6、0.869 3和0.846 2。使用该装置并配合相应的方法能较为准确的一次性测量玉米种粒的三维参数,该研究可为玉米种粒的精细化考种提供参考。
It is an important and difficult problem to acquire and analyze the three-dimensional characteristics of maize seeds after the ear removal. In this study, a maize seed measuring device and the corresponding algorithm were developed based on oblique photography. The device was used to store the spatial information and the calibration data which were obtained from oblique photography. Through 2 mutually perpendicular calibration plates, the three-dimensional data were calculated based on the obtained information and data. The center of the horizontal plate was provided with a circular hole whose diameter is equal to that of the rotating disk（200 mm in the experiment）. The diameter of the round hole was measured as a constant to provide for the system. It provided calibration data for measuring the length and width of maize seeds. The tangent points on the left and right sides and the 4 vertices of the square were clearly marked with red dots. The stepper motor drove the disk to rotate with the speed of 10 degrees per second. The image recording system started with the speed of 10 frames per second and the resolution of 1280×720. With the support of the device, the length, width and thickness of each maize seed were obtained by image processing algorithm. Horizontal and vertical images were taken from oblique photography images by perspective transformation; long axis and short axis of each maize seed were calculated using the diameter of the disk as the reference; the thickness of a maize seed was calculated by taking the calibration data as the reference. Pixel distance measurement method was combined with the watershed algorithm to achieve better image segmentation results. First, global threshold was used to obtain binary image. Then the distance of 2 pixels in the binary image was calculated. At last combined with the watershed algorithm, the boundary of the region was taken as the watershed. A simple and fast calculation method was designed according to the shape characteristics of maize seed to judge the seeds direction. First the centroid was calculated based on the moments of maize contour, and then the pixel whose distance from the centroid was the maximum was taken as the tip of the seed. The direction of the connection between the center and tip point was the direction of the seed. In this study, the length and width of the maize seeds were calculated based on the vertical orthographic images obtained from the perspective transform. Since the centroid position and the tip position of the maize seed contour had been calculated, the length and width of the seeds could be measured based on these data efficiently. After obtaining the back image of a seed in ROI（region of interest）, the thickness was calculated based on the horizontal view from the perspective transformation. The boundary of thickness was obtained according to the horizontal cumulative distribution. An analysis and calculation triangle was established according to the location of a seed. The thickness data were mapped on the checkerboard plane, and accurate measurement results were obtained. Microsoft Visual Studio 2010 was taken as the software development tool, and Open CV machine vision algorithm library was used to develop the experimental program. The experiment was performed with 180 maize seeds selected randomly. The root mean square errors（RMSEs） of the long axis, short axis and thickness were 1.86, 1.28 and 0.741 mm respectively. The determination coefficients of the long axis, short axis and thickness were 0.849 6, 0.869 3 and 0.846 2 respectively. The results show that this device and method can be used to measure the three-dimensional parameters of maize seeds with a relative high accuracy.
出处
《农业工程学报》
EI
CAS
CSCD
北大核心
2018年第4期201-208,共8页
Transactions of the Chinese Society of Agricultural Engineering
基金
国家高技术研究发展计划(2012AA10A501-5)
重庆市教委科技计划(KJ1500320)
关键词
机器视觉
图像处理
测量
玉米种粒
农业装置
倾斜摄影
machine vision
image processing
measurement
maize seeds
agricultural equipment
oblique photography
