Design And Key Technology Research On Detection System Of Droplet Deposition Characteristics Based On Machine Vision

Precise pesticide application,reducing quantity and increasing efficiency is one of the important measures to realize the green development of agriculture in China.With the improvement of the mechanization degree of plant protection in China,the scientific evaluation of spray operation effect of machines and tools has an important guiding role in improving the operation quality of plant protection machinery and reducing the use of pesticides.At present,the application of machine vision technology to spray quality detection is a development trend,but there are some problems,such as low detection accuracy caused by low identification accuracy and poor segmentation effect.In this paper,a portable,highprecision,simple-operated system for detecting droplet deposition characteristics is developed based on machine vision,and key technologies such as the identification and segmentation of adhesive droplets are studied.It solves the problems of expensive instrument,complicated operation and low accuracy in traditional spray detection methods,which has practical significance for realizing rapid detection of spray quality.The research content of this paper mainly includes the following five aspects:(1)Development of droplet deposition characteristics detection device,mainly including hardware platform construction and software system design.To ensure that the whole watersensitive paper image is obtained under the condition of proper structure size of the whole machine,the image sliding acquisition module is designed;Use 3D printing technology to make key parts,assemble and shape the hardware system;Complete the control system and graphic interface design.(2)Sample acquisition and image pre-processing.Water-sensitive paper trays from different angles were designed to obtain various forms of spray droplets falling on the surface of water-sensitive paper.Water-sensitive paper samples were collected through spray operation of plant protection equipment to provide analysis samples for subsequent image preprocessing and adhesion droplet segmentation.In order to improve the accuracy of droplet region extraction during image preprocessing and reduce the interference of factors such as uneven illumination,the results of droplet image extraction by global threshold segmentation,block threshold segmentation and dynamic threshold segmentation are compared;The method of water sensitive paper image pre-processing is determined.(3)Establishment of droplet classification model.Statistical analysis of the shape parameters of different types of droplet images shows that the classification accuracy of single parameter is low.In order to improve the accuracy of adhesive and non-adhesive droplet recognition,k-neighbor,logistic regression,decision tree and support vector machine droplet recognition models were established based on machine learning;Through the comparative analysis of different models,the prediction accuracy of four droplet identification models is 0.94,0.98,0.95 and 0.97 respectively;The optimal logistic regression droplet identification model is selected as the evaluation criterion for the accuracy of adhesive droplet identification and segmentation.(4)Design of adhesive droplets segmentation algorithm.In order to solve the problem of undersegmentation and oversegmentation in the traditional watershed segmentation algorithm,the watershed segmentation algorithm was improved based on the established droplet recognition model.The adhesive droplet in water-sensitive paper image was extracted,and the segmented marked seed point image was obtained through iterative distance transformation processing of the adhesive droplet image,and the segmentation of the adhesive droplet image was completed.The segmentation result is better than traditional watershed segmentation method.(5)System performance test verification.The droplet deposition distribution detection test,droplet size distribution detection test and field test were conducted.Compared with the Deposit Scan droplet analysis software and DJI droplet analyzer,the detection results of droplet coverage presented in this paper were 4.97% and 11.29% higher in accuracy;Compared with the results of manual counting the average error of droplet coverage density was 3.69%;Compared with the results of laser particle size analyzer the average error of droplet size was 8.05%,and the average error of volume median diameter was 6.51%.