Research On Deep Detection Of Uav Droplet Deposition Parameters Based On Network

The deposition parameters of fog droplets are the most important indicators to evaluate the mechanical performance of the quality assurance.For the traditional fog droplet image parameter detection,there are problems such as time-consuming and laborious,cumbersome operations,low detection accuracy of the algorithm,and poor generalization ability.The model of rapid detection of fog droplet deposition parameters on the network has compared and analyzed the accuracy of various fog droplet segmentation algorithms,and optimized the adhesion fog droplet segmentation algorithm to improve the accuracy of the model.This design takes fog droplets as the research object,sprays pesticides through UAV,uses water-sensitive paper in the collection device to collect,the fog droplet image is collected by CCD camera,and the WIFI wireless module is uploaded to the host computer to construct the fog droplet image data set.And based on the traditional image detection algorithm and deep neural network model to model and test the fog droplet image data.The main research contents are as follows:(1)For the fog droplet image segmentation,the maximum internal variance method,the threshold algorithm based on the minimum value of the valley floor,and the adaptive algorithm based on the Wall algorithm are used to binarize the fog droplet image,which is based on the maximum internal variance algorithm combined with Gaussian filtering.3.The median filtering effect of fog droplet image is better.(2)For the segmentation of sticky fog droplets,the comparative analysis of the limit corrosion algorithm,the improved iterative open operation segmentation algorithm,and the improved watershed segmentation algorithm respectively.The accuracy of the region fitting model.(3)For the fog droplet area after segmentation,the fog droplet area fitting model is established based on Hough circle detection,least squares and genetic algorithm,respectively.The least squares circle detection and fitting effect is the best,and the area of fog droplet area s The average accuracy of coverage rate cov and coverage density cov＿dens are 0.886,0.837 and 0.879,respectively.The average detection rate is 1.051 s /frame.(4)For fog droplet images with different coverage densities,the fitting model of the fog droplet area was established based on different depth neural network models.The Faster-RCNN model has the highest accuracy,and the average detection accuracy of the average area s of the fog droplet area,the coverage rate of the fog droplet cov,and the coverage density of the fog droplet cov＿dens are 0.923,0.913,and 0.928,respectively.However,the parameter of the model is very big,and the average detection efficiency is about 1.117 s / frame.Based on the comprehensive analysis of the accuracy of detection of droplet deposition parameters,the amount of parameters of the model and the detection rate as quantitative indicators,the performance of the 0.75＿SSD＿Mobile Net＿262 network model is the best.The average accuracy of the three types of droplet deposition parameters are 0.899,0.873,and 0.901,respectively.The average detection rate is 0.343 s / frame.(5)For mobile terminal detection of fog droplet parameters,the SSD＿Mobile Net model is lightweighted by width multiplier α and resolution multiplier β,and the accuracy and lightness of the model are compromised,α,β Rvespectively 0.75,0.875.In this design,after the lightweight model of SSD＿Mobile Net network is transplanted to Jetson TX2 embedded,the average area,coverage,and coverage density of the fog droplet area are0.894,0.871,and 0.904,respectively.The minimum accuracy rates are 0.88,0.862,and0.893,respectively.The average detection rate is 0.009 s / frame,and the minimum detection efficiency is 0.011 s / frame.