Research On Agricultural Vehicle Operating Environment Perception Technology Based On Machine Vision And Lidar Fusion

In order to improve the working efficiency and intelligent level of agricultural vehicles,it is necessary to realize the effective operating environment perception of agricultural vehicles.The environmental awareness technology of agricultural vehicles needs to be able to operate normally in a very challenging scene,which is open,with sharp changes in light,no structured roads,and many negative obstacles such as ridges,ditches,pits.By analyzing the basic components of convolutional neural network: convolution layer,pooling layer and full connection layer,and referring to the residual structure,the Mini-Res Net10 model is designed.After training and learning,the model can achieve high accuracy in the difficult classification task of sub class fine grain diseases and pests.The m AP for detection of cabbage diseases and pests can reach 94.13% accuracy,and the highest accuracy in the identification of Backmoth pests can reach 95.47% accuracy.YOLOv3-tiny obstacle detection model based on camera image and Point RCNN obstacle detection model based on laser radar point cloud are selected for fusion perception technology research.The transformation equation is constructed to project the point cloud obstacle detection results from the point cloud space to the image space,and the weighted fusion is realized by assigning different weight coefficients to the projection area and the original image.Select the brightness information in the image color feature and the fusion weight coefficient of the original image α build mapping relationships.Assumed weight coefficient α is a quadratic linear relationship with m AP.Get the quadratic fitting relationship between them by interpolation.When the maximum point of the fitting relation is obtained α = 0.73,β = 0.27 weight coefficient.Under this fusion weight coefficient,the fusion model F-YOLOv3-tiny obtained the maximum value of m AP of 76.7%.According to the idea of F-YOLOv3-tiny fusion model,a single line laser radar is designed to locate the negative obstacles in the common ridges,ditches,pits and other negative obstacles on the road of agricultural vehicles,and a method of image negative obstacle segmentation is designed in the located area.In the Gazebo simulator,an orchard farm vehicle operation scene containing potholes is constructed to verify the effectiveness of this method.Simulation results show that this method can improve the segmentation accuracy and speed up the response speed,which can reach the response speed of 365 frames per second.