| In the process of plant pest control,the traditional application method generally uses large-capacity leaching spray technology,which is easy to cause pesticide waste and bring environmental pollution problems.Fine spray on target has gradually become a research hotspot in the field of agricultural and forestry plant protection,which is of great significance for saving pesticides and protecting the environment in plant protection operations.Aiming at the problems of trajectory deviation error in the positioning process of spray robot and the fact that target detection cannot accurately distinguish canopy and trunk,this paper applies trajectory optimization method and plant canopy segmentation method to precise positioning and variable spray target detection of spray robot system,which is conducive to improving the accuracy and refinement of spray application.The specific research content is as follows:(1)A robotic spray system was designed to target the plant spray in the environment of orchard and forest road.The system is mainly composed of three parts: data acquisition unit,motion control unit and spray control unit.The data acquisition unit uses the RGB-D sensor to obtain the color information and depth information of the target plant,and is processed and analyzed by the embedded main processor;The motion control unit uses the gyroscope to obtain the attitude data of the spray robot,and controls the hub motor by the auxiliary controller through the CAN bus to realize the motion control of the robot;The spray control unit uses a diaphragm pump and a PWM control solenoid valve to refine the target spray.(2)Aiming at the problems of traditional trajectory optimization methods such as poor positioning accuracy,floating point drift and easy loss of depth information,this paper proposes a global nonlinear trajectory optimization method that combines depth information.In the process of spraying robot,the color information and depth information of the plant are collected by RealSense sensor.By extracting feature points,calculating descriptors,matching feature point pairs,combining the principle of polar constraint to obtain re-projection error,and integrating depth information iteration Reprojection error to obtain accurate spray robot travel trajectory.The experimental results show that the proposed method effectively reduces the dependence of the singledepth information in the optimization process of the spray robot trajectory,so that the average deviation of the fitted trajectory of the spray robot is reduced by 1.07 cm,the variance is reduced by 2.14 cm,and the overshoot is reduced by 3.13 cm,greatly improving the accuracy of the spray robot trajectory estimation.(3)For the traditional super-voxel clustering method in plant detection and segmentation,the over-segment rate is high and the real-time performance is poor.This paper proposes a super-voxel clustering method that combines significant point clouds.The method converts the color information obtained by the Kinect 2.0 sensor into a saliency feature map,and synchronously aligns the depth information to obtain a saliency point cloud,and then the improved super-voxel clustering method and the LCCP algorithm cluster twice to obtain a complete segmentation image.The experimental results show that the improved saliency point cloud super-voxel clustering method effectively overcomes the influence of background noise and outliers on segmentation,and its over-segmentation rate is lower than that of traditional supervoxel clustering method by 1%~3%.The time of clustering of super-voxels decreased by 3.43 s,which improved the accuracy and rapidity of canopy and trunk segmentation of spray target plants. |
