Research On Visual Navigation Path Recognition Of Field Tillage Weeding Robot In Complex Environment

With the development of artificial intelligence and machine vision technology,visual navigation line extraction has gradually become a core technology of autonomous navigation of agricultural robots.Aiming at the complex and changeable unstructured environment in the field,this paper conducts a systematic study on the extraction of ridge navigation lines between field and greenhouse crops based on machine vision and image processing.The main contents and conclusions of the study are as follows.(1)The navigation line extraction of the intertillage weeding robot based on visual navigation is susceptible to weeds,variable light,lack of plants and broken rows during the seedling and mid-growth stage of the crops.This paper proposes a method of navigation line extraction based on the Hough transform of median point for maize rows detection.In order to solve the problem of time-consuming and low accuracy of machine vision in extracting the navigation line of early maize seedling rows under multiple environmental variables,the ultra-green grayscale algorithm was improved in the image pre-processing stage,and the background segmentation of soil and crop seedling rows was realized by using the maximum inter-class variance method and mathematical and morphological operations.Furthermore,the feature points of the crop rows were extracted by the mean value method,and then the maize seedling lines on both sides of ridges were fitted according to Hough transform of median points.Finally,based on the maize seedling lines on both sides,the navigation lines were extracted by using the tangent formula of the included angle.The experimental results show that the maximum error of the navigation line extracted by the improved line detection algorithm is 0.53°,which is on average 62.9 ms faster than the standard Hough transform time.After verification and statistics,the accuracy of navigation line extraction has reached more than 92%,and the proposed algorithm has good accuracy and reliability.(2)In order to avoid the problem of seedling damage of robots in field operations,the Faster-U-net model was proposed based on the U-net model.By collecting the crop row data sets in the field and greenhouse and training them based on the Faster-U-net network,the prediction and identification of the paths between ridges of crops were realized.Further,the path navigation lines of the robot were extracted based on B-sample curves.The experimental results show that the recognition accuracy of the Faster-U-net network model for maize,tomato,cucumber and wheat is 93.86%,94.01%,93.14%,and 89.10% respectively.The average angular difference of the extracted navigation lines in the tillage environment for the crops of maize,tomato,cucumber and wheat is 0.624°,0.556°,0.526° and 0.999° respectively.The improved algorithm has strong robustness and accuracy.(3)According to the actual field operation scene of the weeding robot,the chassis structure of the intertillage weeding robot was designed through 3D software,which was suitable for the field navigation control of the robot.And then the robot motion control model based on differential steering was analyzed,and the fuzzy PID control was compared with the traditional PID control.The experimental results show that the effect of the fuzzy PID control is better than the traditional PID control.The designed mobile platform of the weeding robot has a good application prospect.This study can provide technical references for the research and development of intelligent agricultural robot navigation equipment in this field.