Design And Research Of Spray System For Jujube Orchard In Southern Xinjiang Based On Machine Vision

At present,the spraying equipment for dwarf and densely planted jujube orchards in southern Xinjiang is always in the open state during the spraying process.It cannot accurately spray the jujube trees of different heights.environmental pollution,etc.In addition,although some devices use image recognition technology,they cannot accurately extract features due to hardware resources,etc.,and the recognition accuracy and efficiency are low,making it difficult to meet the requirements of precise spraying.In this paper,aiming at the dwarf and densely planted jujube orchards in southern Xinjiang,on the basis of analyzing the research on plant protection spray technology based on machine vision at home and abroad,two key points are formulated for the real-time identification and classification of the Raspberry Pi development platform and the spraying rules for dwarf and densely planted jujube orchards in southern Xinjiang.Research on the technology and propose a spray algorithm based on Re-Se-Inception.And formulate relevant evaluation indicators to test and evaluate it.The main research contents of this paper are as follows:（1）The overall structure of the system is designed,and the core components are selected and determined according to the actual needs to achieve the purpose of overall compactness and high spray quality.（2）Aiming at the problem of difficult image acquisition,a small dataset creation method based on a fixed scene is proposed,and a dataset of dwarf densely planted jujube trees is created by this method.This paper selects the common convolution kernel stacked CNN model,the Inception NET model designed based on the Inception module,and the Re-Se-Inception Net model.Then,under the same host environment,the three models are trained on the dataset of this paper using the same hyperparameters for comprehensive comparison.From the results of the recognition and classification experiments,it can be concluded that the Re-Se-Inception Net model designed in this paper has a recognition and classification accuracy rate of 93.32%,with a standard deviation of 0.18.It meets the deployment requirements for real-time detection on the Raspberry Pi4B development board.（3）The probabilistic trust mechanism and behavior prediction mechanism are combined in the Re-Se-Inception algorithm to form the spray algorithm of this paper.Implemented on the Raspberry Pi platform using the Tensor Flow framework and Python language.（4）A machine vision spray system based on Raspberry Pi4B is designed.In view of the lack of unified evaluation indicators in the industry,three evaluation indicators were formulated for the spray control algorithm:the real-time rate p_rt,the missed spray rate p_mr and the pesticide application reduction rate p_rsr.In the experimental environment of the dwarf and densely planted jujube orchard in the eleventh group of Alar,the real-time rate p_rt is tested first.The experimental results show that the application effect of the spray control algorithm is verified by the experiment.First,the real-time rate p_rt is tested.According to the obtained experimental data,there is a negative correlation between p_rt and the device movement rate.When the movement rate is equal to 0.3m/s,p_rt=91.7%.Then,the reduction rate p_rsr and the missed spray rate p_mr were further tested.The fixed moving rate was 0.3m/s.It was found that when p_mr and p_rt were 10.3% and 91.7%,respectively,p_rsr was 44.5%.It can be seen that A good reduction effect was achieved.