基于SLAM与深度学习的植保机导航系统设计

Design of Plant Protector Navigation System Based on SLAM and Deep Learning

随着智能植保机的广泛应用,其自主定位与导航要求更高,PS/INS组合导航系统存在遮挡物导致信号弱、民用型导航偏差较大局限性,基于激光的自主定位建图技术(SLAM)在现代农业大棚中有精度高、灵活性好等优点,但激光SLAM技术存在激光点云扫描不能准确区别目标农作物和其他障碍物,从而难以保障植保机运行轨迹的正确性。提出了基于激光SLAM技术、ROS系统融入深度学习的目标检测算法,以正确区分扫描目标物体,从而为基于SLAM技术的植保机提供了更为有效的导航方法,提高其工作效率。

With the widespread application of intelligent plant protection machines,higher autonomous positioning and navigation are required,PS/INS integrated navigation system is limited due to the presence of occlusion,which leads to weak signal and large deviation of civil navigation.Laser-based autonomous localization mapping technology(SLAM)has the advantages of high precision and good flexibility in modern agricultural greenhouses.However,laser SLAM technology exists that target crops and other obstacles cannot be accurately distinguished by laser point cloud scanning,so it is difficult to guarantee the correctness of plant protector's running track.In this paper,a target detection algorithm based on laser SLAM technology and ROS system combined with deep learning is proposed to correctly distinguish the scanned target objects,thus providing a more effective navigation method for plant protectors based on SLAM technology and improving their working efficiency.