Research On Key Technologies Of Plant Protection Spray Based On Image Recognition

At present,most of the plant protection spray equipment working in the intercropping farmland environment does not have the image recognition function.The nozzles are always opened during the whole process of the spraying,which causes waste of pesticides and pollution of the soil environment.The rest few plant protection spray equipment with image recognition function use relatively backward image recognition technology due to the limitation of the hardware performance of the equipment,which have the defects of difficult feature extraction and low recognition accuracy and speed.In order to solve the above problems,this paper conducts more specific research on two key technologies including real-time recognition on embedded platforms and formulating spray rules for intercropping farmland after analyzing the researchs on plant protection spraying technology based on image recognition,then designs an SBY（Spraying Based on Yolov3）spray algorithm based on improved Yolov3,and formulates evaluation indicators for it.The main research contents and innovations of this article are as follows:1)Aiming at the shortcomings of traditional image recognition technology,a method of object detection technology based on convolutional neural network was proposed.The current mainstream Two Stage and One Stage object detection algorithm were systematically studied,combined with the plant protection spray operation requirements,the Yolov3algorithm was selected as the basic algorithm of this article.2)A method of making small data sets based on flowing scenes was proposed,and a peperomia tetraphylla plant data set was made by using this method to train the Yolov3model in the host environment.The number of frames detected by the trained model on the NVIDIA Jetson TX2 embedded AI platform was only 3 frames,which cannot meet the real-time requirements.In response to this,the Yolov3 algorithm was improved based on longitudinal depth compression and lateral channel pruning.The improved Yolov3 algorithm was 79.4%smaller than the original model,the detection accuracy was improved by 11.2percentage points,and the number of detection frames was increased to 26 frames,which meets the deployment requirements for real-time detection on Jetson TX2 platform.3)Through the analysis of the droplet movement process during the plant protection spray operation,the spray rules under static conditions were formulated;for the defect that the single threshold mechanism is not flexible enough in actual work,a probabilistic trust mechanism based on target confidence was proposed;to deal with the spray delay problem caused by the number of frames per second lower than the number of frames collected by the camera,a behavior prediction mechanism based on the current vector speed of the device was proposed.Then,the improved Yolov3 algorithm,static spray rules,probabilistic trust mechanism and behavior prediction mechanism were merged to form the SBY（Spraying Based on Yolov3）spray algorithm,and the algorithm was implemented on the NVIDIA Jetson TX2 platform based on the Darknet framework and C++language.4)A plant protection spray equipment platform based on NVIDIA Jetson TX2 was built,and the simulated peperomia tetraphylla intercropping farmland was designed,thus forming an indoor simulated experiment environment.In view of the current situation of the industry lacking unified evaluation indicators,combined with the artificial spray effect and the expected requirements of the project,three evaluation indicators of real-time rate,leakage spray rate₈4))and pesticide reduction application rate_{（90）（8） were formulated for the SBY spray algorithm.In the indoor simulated experimental environment,the real-time ratewas first tested.The experimental results show that the real-time rate of the SBY spray algorithm decreases with the increase of the plant protection spray equipment moving speed.When the equipment moving speed is 0.3m/s,reached 93.3%.Maintain the equipment moving speed at 0.3m/s,and conduct correlation experiments on the three indicators in the same farmland environment.The experimental results show that the（90）（8）reached 47.4%when thewas 89.7%and the84}was only 9.3%,the performance of SBY algorithm in reducing pesticide application was good.