Research On Navigation Technology Of Automatic Weeding Robot In Orchard

Orchard weeds breed pests and absorb nutrients from fruit trees,resulting in a reduction in the yield of economic crops in orchards,while manual weeding and chemical weeding will consume a lot of human,material,and financial resources and will cause certain damage to the environment.With the development of computer technology,it has become a new trend to use intelligent agricultural machinery to replace humans in weeding,and autonomous navigation technology is the core of realizing automatic weeding by agricultural machinery.The selection of navigation reference objects in the natural orchard environment without weeding becomes difficult,and the interference in the natural environment is complex and numerous.According to the characteristics of orchards in the natural environment and the relevant requirements of the topic for automatic weeding robots,this paper obtains the navigation centerline of the orchard by studying the target recognition and positioning method of the orchard navigation reference object in the natural environment,and constructs the two-dimensional coordinate information of the orchard.Simple orchard map,and then design the corresponding fuzzy control algorithm according to the obtained navigation parameters,and finally design and build the navigation system software platform,and carry out related experiments in natural orchards.The main research contents of this paper are as follows:1)Collect 400 pictures(1200 pictures in total)of the test orchard under three different lighting conditions of normal light,high light and mixed light through the binocular camera,to create a training data set of fruit tree trunks,and use transfer learning method to analyze them.At the same time,the Yolov4 target detection algorithm based on convolutional neural network was used to construct the recognition model of fruit tree trunks,and the orchard pictures under three kinds of lighting were recognized.The test results show that the average recognition accuracy of the model trained in this paper is95.38%.2)Establish the mutual transformation relationship between the world coordinate system,camera coordinate system,image pixel coordinate system and image physical coordinate system four important coordinate systems for camera imaging,and on this basis build an ideal binocular imaging ranging model,using Matlab The toolbox uses calibration method proposed by Zhang Zhengyou to perform camera calibration experiments on the left and right eyes of the binocular camera,and solves the internal and external parameters of the left and right cameras;the position and attitude of the weeding robot are solved by using geometric methods,and the lateral offset distance and deflection are calculated.angle as a navigation parameter.3)By comparing the time-consuming and matching accuracy of the BM algorithm,GC algorithm,and SGBM algorithm,the SGBM algorithm is selected to perform stereo matching on the images containing the feature information of the navigation target,and combined with the Opencv vision library and the binocular camera in the Microsoft Visual Studio 2016 development environment The calibration parameters obtained the depth map of the relevant image,and then completed the mapping from pixel coordinates to 3D coordinates through Reproject Image To3 D.The average deviation rate of the target feature positioning on the X and Y axes was 2.21% and 2.21%.Finally,a map containing the coordinate position information of the fruit trees is simply constructed.4)According to the operation requirements of the automatic weeding robot,combined with the navigation parameters obtained from the solution,the motion control method is studied.The fuzzy control rules are designed in Matlab,and the corresponding fuzzy controller is generated.Finally,the fuzzy control algorithm designed in this paper is carried out.Experimental simulation,the experimental results show that the fuzzy control algorithm designed in this paper can meet the navigation requirements.5)In order to realize the mutual communication between the various modules of the automatic weeding robot,a navigation software system was built in the Windows system based on Qt and Opencv integrated in Microsoft Visual Studio2016 using C++ language,and the navigation performance of the automatic weeding robot was carried out in the natural orchard.The experimental results show that when the mobile chassis travels at the speed of 0.2m/s,0.3m/s and 0.4m/s under three different lighting conditions,the total average offset distances are 2.87%,3.36% and 4.14%,respectively.The method studied in this paper can meet the actual navigation needs of automatic weeding robots under different speeds and light intensities.