基于深度学习算法的采摘控制系统高效化设计

Efficient Design of Picking Control System Based on Deep Learning Algorithm

设计了一款基于履带运动的苹果采摘机器人控制系统。工作原理为:首先,通过安装在机械臂上的工业摄像机采集苹果图像,发送给总控制系统,总控制系统利用深度学习-YOLOv3算法识别目标苹果,并检测出其位置信息;然后,由安装在夹爪上的测距传感器采集苹果与夹爪之间的距离信息,进行位姿解算,并将结果发送给机械臂运动控制卡,运动控制卡发出动作指令,使得机械臂到达目标采摘位置,夹爪末端的摄像头、红外传感器会辅助确认苹果落入夹爪内;此时,末端执行控制模块发送抓取指令,控制步进电机运转,使得夹爪抓取苹果,并通过压力传感器反馈夹爪闭合的程度,以免抓伤苹果;最后,末端执行控制模块驱动切片电机切割树枝,摘取苹果。实验结果表明:使用该方法准确采摘苹果的概率达到86.67%,基本能够满足苹果采摘作业的需求。

This paper designs a control system of apple picking robot based on crawler motion. The working principle of the system is as follows: firstly, the apple image is collected by the industrial camera installed on the manipulator and sent to the general control system. The general control system uses the deep learning yolov3 algorithm to identify the target Apple and detect its position information;Then, the distance sensor installed on the clamping claw collects the distance information between the apple and the clamping claw, performs pose calculation, and sends the result to the manipulator motion control card, which sends action instructions to make the manipulator reach the target picking position, and the camera head and infrared sensor at the end of the clamping claw will assist in confirming that the apple falls into the clamping claw;At this time, the end execution control module sends the grasping command to control the operation of the stepping motor to make the clamping claw grasp the apple, and feed back the closing degree of the clamping claw through the pressure sensor to avoid scratching the apple;Finally, the end execution control module drives the slicing motor to cut branches and pick apples. The experimental results show that the probability of accurately picking apples by this method is 86.67%, which can basically meet the needs of apple picking.