Design And Experiment Of Field Weed-targeted Spraying Device Based On Machine Vision

In the past decade,the average annual usage of chemical pesticides in China exceeded 1.5 million tons,while the average effective utilization rate was only30% to 40%.Most pesticides in traditional indiscriminate coverage spraying methods were unable to intercept the target,eventually entering the soil and environment,causing serious waste and pollution.To address these issues,this paper developed a real-time target spray device based on machine vision for field weed control scenarios.The device adopts a systematic and modular design,deeply integrating deep learning object detection technology,electronic control technology,and agricultural spraying machinery.By accurately identifying field weeds and spraying targeted pesticides,the amount of pesticides used can be significantly reduced.The specific research content includes:(1)According to the requirements of targeted spraying tasks,the overall scheme of the machine vision-based targeted spraying device was designed,including four modular hardware parts: image acquisition and recognition module,electronic control spraying module,stable drug supply module,and power supply system,and two core control parts: deep learning weed detection model and accurate spraying control algorithm.According to the design scheme and indicator model,the selection and optimization of each component were completed.(2)Taking common northern weeds such as prickly lettuce as an example,a lightweight deep learning field weed detection model was constructed.To solve the problem of low efficiency of manual field image acquisition,a highly convenient remote field image acquisition car was designed and produced,and 3200 weed images were collected in multiple plots and field environments to establish a high-quality weed dataset.Based on the commonly used object detection model YOLOv5,a lightweight object detection improvement model YOLOv5-Mobile Netv3-SE was designed and developed by replacing the feature extraction backbone network and adding an attention mechanism.Compared with the YOLOv5 s original model,the model size was reduced by 53.5% under the condition of basically no decrease in detection accuracy,and the detection frame rate was increased by 27.8%.The main performance indicators were: m AP@0.5 was86.9%,the model size was 7.5 MB,and the FPS was 38.17,which meets the real-time targeted spraying task requirements.(3)An accurate targeted spraying control algorithm with real-time delay compensation function was designed.The algorithm adopts a spraying strategy based on impact grid matching,converts weed position prediction information into spraying information,and controls the start and stop of the nozzle spraying according to the starting position of the weed canopy.By measuring the fixed delay of each subunit(27.90 ms for image processing,6.37 ms for communication control,and 36.40 ms for spraying deposition)and the real-time traveling speed of the device,the advance amount is predicted and the position of the matching grid is corrected to eliminate the lag of targeted spraying caused by system execution delay,ultimately achieving accurate spraying control.(4)A performance evaluation test bench was built to complete comprehensive performance evaluation testing of a machine vision-based target-oriented herbicide spraying device.The results showed that at speeds of 1 to 4 km/h,the spray hit rate of this device on the weed model was above 85%,and the hit rate showed speed dependence: the spray hit rate decreased with the increase of the speed.Compared with traditional spraying,the drug-saving rate of target spraying was 48.6%,55.8%,72.5%,and 79.0% at weed coverage rates of 20%,15%,10%,and 5%,respectively.(5)Field tests of the target-oriented herbicide spraying device based on machine vision were completed,and the recognition model demonstrated high adaptability to complex scenes.At speeds of 2 km/h,3 km/h,and 4 km/h,the target spray hit rates were 90.80%,86.20%,and 79.61%,respectively.In summary,the target-oriented herbicide spraying device based on machine vision developed in this paper achieves accurate identification of weed targets in the field and precise pesticide spraying,significantly reducing pesticide pollution during weed control operations,and providing technical support for green and sustainable field production.