Research On Cross-row Self-propelled Intelligent And Efficient Weeding Robot Adapted To Multi-terrain

Weeds in farmland can seriously affect crop yields and quality,so field weeding is a top priority.The current weeding methods include manual weeding,chemical weeding,mechanical weeding,biological weeding and thermal weeding,in contrast,mechanical weeding has the advantages of environmental friendliness,high efficiency and low artificial intensity of farmland,and also has the advantages of improving the soil environment and enhancing soil permeability.Based on the existing domestic and foreign weeding equipment,this paper proposes a cross-row self-propelled weeding robot adapted to multi-terrain,which is equipped with inter-row weeding device,inter-plant weeding device and identification system and its control system,which can remove weeds between rows and plants at the same time,and identify field crops to avoid crops.The main research contents and conclusions of this paper are as follows:(1)The specific scheme design of cross-row self-propelled weeding robot was carried out.Taking pepper seedlings as the research object,the agronomic characteristics of planting,the technical requirements of weeding and the specifications of their ridge fields were analyzed,and the design of the mobile platform of the device was improved.After designing the inter-row weeding device and the inter-plant weeding device,and the three-dimensional model was established,the height adjustment device and the row spacing adjustment device were designed on it,so that the device could adapt to different ridge fields.The inter-row weeding device uses a weeding shovel to remove weeds between the rows,and the inter-plant weeding device uses a weeding knife mounted on a crank rocker device to remove weeds between plants,and its movement principle allows it to avoid crops.Secondly,the motion analysis of Soildworks software is used to simulate and verify the reliability of its motion.The motor required for each device is also selected for calculation.(2)Discrete element simulation analysis of weeding device.The resistance of the soil and the disturbance of the soil by the tool are studied by discrete element simulation method In EDEM software,the model of the weeding tool and the soil is designed,and the required parameters are set and the simulation analysis is carried out.The simulation results show that when the travel speed and weeding depth are fixed,the force law of the weeding tool between rows and plants decreases first and then increases with the increase of the soil entry angle.At the same time,compared with the maximum force and amplitude change,the inter-row weeding device selected 45° soil entry angle as the optimal working parameter,and the interplant weeding tool selected 60°soil entry angle as the optimal working parameter.Among them,the inter-plant weeding knife has less disturbance to the soil and does not need to considered;After the operation of the inter-row weeding cutter,the soil fluffiness of the inter-row weeding tool is between 10%and 40%,and the soil disturbance coefficient is greater than 50%,which can realize the disturbance of the surrounding soil and will not cause disturbance damage to the surface structure.(3)A network model for pepper seedling recognition based on deep learning was studied.The structural characteristics of convolutional networks are studied and analyzed,which have the characteristics of high self-learning ability,strong adaptability and high efficiency.First,the dataset required for the model was made,images of pepper seedlings were collected under different weather conditions and different time periods,and data augmentation was adopted to increase the richness and effectiveness of the dataset.After the typical recognition algorithm is used for research,there are YOLO series algorithm,SSD algorithm,Faster R-CNN algorithm,and the model is trained,and the results of the training and the actual effect of its prediction are compared and analyzed,and the results are as follows:Among the four model prediction results,the performance performance of YOLOv5 stands out in these algorithm models,its accuracy is 89.98%,the recall rate is 96.59%,the F1 score value is 0.93,and the mAP value is 96.21%.The real-time detection speed is 34.42FPS and the detection speed of a single image is 29.05ms,which generally meets the requirements of high recognition accuracy and real-time performance required by cross-row self-propelled weeding robots in the field.(4)Produce a prototype of a cross row self-propelled weeding robot and conduct field experiments.Processing,installation and commissioning are carried out according to the design scheme and three-dimensional model of the weeding device.For the previous discrete element simulation results,the field tool test is carried out,and the film pressure sensor is used to measure the actual force value,and the results indicate that the error between the real field tool force detection result and the discrete-element simulation result is within 20%,which proves its reliability and the feasibility of the weeding device.Then,the traveling speed of different levels and the depth of soil entry of the weeding knife were selected as the influencing factors,and the weeding rate and seedling injury rate were used as the evaluation indicators to study the performance of cross-row self-propelled weeding robot in the field,and the experimental results were as follows:the primary and secondary orders affecting the weeding rate were weeding depth>marching speed,and the primary and secondary orders affecting the seedling injury rate were marching speed>weeding depth.Through comparative analysis,the optimal working parameter of 0.5m/s forward speed,the weeding depth between rows is 30mm,and the weeding depth between plants is 6mm,which meets the requirements of weeding rate greater than or equal to 85%and seedling injury rate is less than 5%,which meets the requirements of weeding performance.