Walking Path Planning Of Maize Growth Information Collection Robot

With the increasing maturity of artificial intelligence and image technology,agricultural machinery has begun to develop in the direction of intelligence,and more and more scientific research institutions and enterprises have launched research on agricultural robots.Image recognition and deep learning have begun to be widely used in large-scale machinery such as tractors and harvesters.Due to the narrow spacing between rows of corn,there are branches and weeds obstructing the path between rows in the middle and late stages of corn growth,and insufficient light makes it difficult for ordinary machines to walk between t he rows of corn and cannot understand the growth of corn in real time.In order to be able to understand the corn crops in real time Growth status,design a kind of corn interrow information collection robot.According to the theoretical basis of convolutional neural network algorithm and related research on the application of existing image recognition technology in navigation technology,the path planning method of crop growth information collection robot based on Yolo v4 is proposed.The specific contents of the research are as follows:First of all,this paper designs an information collection robot that can walk between rows of corn.The robot is driven by a hub motor,a stepper motor controls a rack and pinion for steering,and a K inect v2.0 camera per forms image collection,field information collection,and comparison.The robot performs positioning,image processing and navigation control through the industrial computer,and the inertial measurement unit collects the posture information of the robot d uring walking.Then,the related theoretical knowledge of convolutional neural network and the principle of coordinate system conversion are introduced.The camera is calibrated according to Zhang Zhengyou’s checkerboard calibration method.After that,the pictures of the rhizomes of each period of corn growth are collected regularly and the pictures are expanded to construct a corn rhizome sample library.Next,frame selection and calibration of the corn rhizome data set,generate a data set file,use the Yolo v4 network model to train the corn rhizome pictures,evaluate the prediction effect and verification results through the model evaluation criteria,and predict the results with Faster R-CNN For comparison,it is found that the accuracy of Yolo v4 is 10.48% higher than Faster R-CNN.Then,the walking path of the robot is fitted according to the recognition result,and the pixel coordinates of the picture are converted into space coordinates through coordinate conversion to obtain the walking path of the robot in the actual field environment.Finally,experiments were carried out in a simulated environment on the ground and in the field,and the experiments showed that the crop growth information collection robot based on this method walked stably and did not appear out of control.The robot can be well applied to corn rows,collect various information during corn growth,prevent and control corn diseases and insect pests in the middle and late stages,and realize the separation of man and machine.