为了满足穴盘苗自动化分选的实际需求，该文设计了基于线结构光视觉的穴盘苗外形参数测量系统，实时获取穴盘苗图像信息，实现对其叶片面积和高度的在线测量。为充分突显目标与背景色彩差异，针对穴盘苗叶片和背景基质图像特征，利用最大类间方差动态阈值对2G-R-B 色差图像进行分割；以穴孔为单位进行区域标记和特征提取，分别计算幼苗叶片图像面积，排除明亮蛭石颗粒造成的椒盐噪声和劣苗叶片区域；根据 Cb、Cr 色彩分量特征提取在健康幼苗叶片区域的红色激光条像素坐标，拟合其分布中心线；基于线结构光视觉三维定位原理，根据幼苗叶片区域激光条中心线图像坐标，实现对穴盘苗高度的测量。试验结果表明，系统对直立姿态的穴盘苗高度测量精度为5 mm，在叶片面积测量评估方面可以满足穴盘苗筛选精度要求。%The tray-seedlings used for mechanical grafting and transplanting should be as uniform as possible, and the tray holes with either nothing or a bad seedling should be rejected. However, it is hard and costly work to pick seedlings from the tray by human choice. In order to meet the need for the tray-seedling’s automatic grading before the mechanical transplanting and grafting, a new system for measuring the seedling feature was designed based on the structured-light vision technology, which could get the leaf size and the height of the seedling through on-line detection. The system was supposed to serve the seedling grading machine. Two color images of each seedling line in the tray were taken by a camera, and the one without linear light was used for identifying leaf size, with the other with linear light used for measuring seedling height. As the major background in the seedling color image, the gray of the soil was varied from its different moisture and mixing-ratio. So the calculation 2G-R-B of the chromatic component was used to distinguish the seedling leaf from the substrate, and the Otsu dynamic threshold was adopted to extract the leaf area. The huge amounts of noise pixels were still left in the binary image, because of the roseite particles appearing outstandingly bright in the soil, In order to clean the noise from the roseite, the white area in every tray hole was labeled sequentially, and counted separately. The area containing more than 4,000 pixels was considered as the seedling leaf, and if not, the area was considered as the noise, bad seedling, or non-seedling. The pixel numbers represented the seedling leaf size, according to which the tray holes with the smaller leaf or non-leaf were identified. The calibration for the linear vision system was completed through processing 20 images of the chess-shaped checkboard. The images without linear-light were used to calibrate the internal parameter, and those with linear light were used to get the external parameter of the structured-light vision unit. According to the linear structured-light vision principle, the 3D coordinate of the light-line on the seedling leaves could be obtained, when the image pixels of the light line were extracted. Besides, the XY plane of the coordinate system was built on the seedling tray, so that the seedling height was same with the coordinate value Z. The pixels of the linear light of 650nm wavelength lying in the leaf area were acquired through the threshold of Cr (97,137) and Cb (82,132), a center line of the light pixels was drawn, and then the coordinate Z of three points in the center line were measured, among which the maximal one represented the seedling height. As the result showed, this method can exactly evaluate the leaf size and the seedling height to satisfy the demand on the automatic seedling classification, and the height measure error is less than 5mm for the normally straight seedling.
展开▼
