Three-Dimensional Recognition of Agricultural Crop

本研究的目的是开发一种三维视觉系统来识别农作物，实现农业自动化。新鲜市场的黄瓜，在大小适合市场时，由人工选择性采收。由于果实生长迅速，收割工作必须经常进行。为了实现黄瓜收获工作的自动化，我们制造了一个用于试验目的的3d视觉系统，并将其安装在黄瓜收获机器人上。3d视觉传感器发射红外激光束。光束被扫描镜反射到农作物上。作物表面的反射光返回到3d视觉传感器。此外，它被另一个扫描镜反射，并通过透镜聚焦在PSD（位置敏感装置）上。通过PSD的两个阳极的信号计算到每个扫描点的距离。像素数为7500，即水平为60，垂直为125。实验是用黄瓜模型进行的，这些模型是根据真实的黄瓜制作的。水果的位置和大小可以相当精确地测量出来。还可以识别出脚梗的位置。三维视觉传感器被认为对黄瓜收获机器人是有效的。采用后续栽培法对黄瓜进行了试验栽培。(1)斜架训练(2)沿斜杆训练(3)保持茎部倾斜但与作物行平行的挂茎训练（3号柱栽培的黄瓜果实挂在茎上最明显，因为果实挂在茎上，被叶片遮挡的概率小）。认为该训练方法适用于黄瓜收获机器人。对脆头生菜进行了识别试验。通过对三维图像的处理，可以识别出脆头的大小。利用安装在末端执行器末端的三维视觉传感器对樱桃番茄进行识别。随着末端执行器的运动，可以从不同的角度扫描识别出许多水果（包括隐藏在叶子中的水果）。少

The objective of this study is to develop a 3-D vision system to recognize the crop for the automation of the farm work.Cucumber for the fresh market is harvested selectively by the human labor when the size is suitable for the market. As the fruits grow rapidly, the harvesting job must be done frequently. To automate the cucumber harvesting job, a 3-D vision system was manufactured for the trial purpose and was mounted on a cucumber-harvesting robot. The 3-D vision sensor emitted an infrared laser beam. The beam was reflected by a scanning mirror toward the crop. The reflected light from the crop surface returned to the 3-D vision sensor. Further, it was reflected by another scanning mirror and was focused on a PSD (Position Sensitive Device) by a lens. The distance to each scanning point was calculated by the signals from two anodes of the PSD.The number of pixels was 7500, that is, 60 in horizontal and 125 in vertical. Experiment wis carried out using cucumber models that were made … More based on a real cucumber. The location and the size of the fruit could be measured with fairly good accuracy. The location of the peduncle could also be recognized. The 3-D vision sensor was considered to be effective for the cucumber-harvesting robot.Cucumber was cultivated by trial in following training method.(1) Inclined trellis training(2) Training along the inclined pole(3) Training by hanging the cucumber stalk keeping the stalk inclined but parallel to the crop rowCucumber fruits on the stalk that was cultivated in the style of No.3 was most visible because the fruit hanged from the stalk and the hiding probability by leaves was small. The training method was considered suitable for the cucumber-harvesting robot.Recognition experiment of the crisp head lettuce was carried out. The size of the crisp head could be recognized by processing the 3-D image.Recognition of the cherry tomato was tried by a 3-D vision sensor mounted at the end of end effector. Many of the fruits (including the fruits hidden by the leaf) could be recognized by scanning from various view points with the motion of the end effector. Less