Research On Unmanned Operation Technology Based On Multi-features Of Rice And Wheat Harvest Scene

In recent years,with the aging population and the hollowing out of farmers in China,there is a serious shortage of personnel who can skillfully drive combine harvesters.Unmanned technology for combine harvesters is developing rapidly,but the existing technology is still far from adapting to the small and irregular field conditions commonly found in China.In this paper,the research on the unmanned operation technology based on the multi-features of rice and wheat harvesting scenes is carried out.The extraction method of field navigation information and crop lodging feature information based on 3D point cloud and the unmanned harvesting control algorithm are proposed.And carried out precision verification and field actual harvest test.The research and work in this paper are as follows:(1)Aiming at the needs of combine harvesters for unmanned harvesting operations based on scene multi-feature feedback,this study designed an unmanned information acquisition scheme with a combination of RGB-D camera and electronic compass.Research on the truly unmanned design of combine harvesters,the scene information ji perception of the cutting line,field head and fallen area,to achieve automatic control of the body position,harvesting path and height of the cutting table.At the same time,to the pressure of real-time calculation brought by navigation information processing and the possible mutual interference when multiple data are transmitted simultaneously,the ROS system with distributed network is built as the upper computer software environment,and the Modbus communication protocol is used to realize the communication between the upper computer Jetson TX2 development board and the lower computer PLC.(2)In order to obtain combine harvester field navigation information,a cut line segmentation algorithm was designed based on 3D point cloud data for the feature that the harvested and unharvested areas of mature rice and wheat are similar in color but differ significantly in height.The point cloud of the cut line area is extracted by random sampling consistent plane fitting and radius neighborhood search,and the cut line linear equation is fitted by the least squares method to calculate the heading deviation and cross-line deviation information.(3)In order to obtain combine harvester field head steering information,a far and near field head line fitting method is proposed based on 3D point cloud data for the high-low-high characteristics between the field head,ground and crop.The heightmajority points in the point cloud are obtained by scanning the Y-direction depth image created from the 3D point cloud column by column,and the far field head line is fitted using an improved random sampling consistency algorithm.Then,we search the near field head area in the direction of the origin of the camera coordinate system,translate the far field head line to pass the near field head area and obtain the equation of the near field head line to calculate the vertical distance information between the combine harvester and the near field head line.(4)In order to realize the unmanned harvesting operation based on the characteristic feedback of the combine harvester on the lodging area,and for the feature that all fallen areas are located under the plane composed of unfallen areas,the segmentation of lodging area and the comprehensive extraction algorithm of positiondegree-direction feature information are designed based on 3D point cloud data.The uninvested plane is fitted by a random sampling consistency algorithm that translates and rotates the point cloud so that the camera coordinate system falls on top of the plane.Finally,CSF ground point filtering and Euclidean clustering were used to extract the point cloud of the inverted area,and information on the location,degree and direction of the inverted area was obtained by coordinate traversal,mean height calculation and surface normal estimation.(5)Field validation trials were carried out and the results showed that the average error of heading is between 0.04° and 1.48°,the average error of lateral is between1.37 cm and 2.53 cm,and the average error of vertical distance from the field head is between-0.034 m and-0.019 m.Then,the average error in the mean height of the fallen area was between 0.004 m and 0.092 m,and the average error in the fall direction was between 0.34° and 9.17°.The accuracy of the system is verified through actual field harvest experiments under different lighting conditions.The results show that the average tracking error of the secant line is between 1.15 cm and 2.19cm;the average steering distance error in the field head is between 0.027 m and 0.082 m,and the average steering angle error is between 0° and 3°;the average error of header height is between1.36 cm and 4.57 cm when the lodging area is harvested.