Droplet deposition characteristics detection method based on deep learning

Abstract

Accurate acquisition of spraying quality parameters of plant protection Unmanned Aerial Vehicle (UAV) is helpful to analyze the deposition and distribution of pesticides in field crops. It is of great significance for pest control. UAV spray quality detection system was used to collect droplets, the droplet deposition parameters can be detected by processing the droplet images. However, the spots and the adhesion of the droplets led to large data errors. In order to solve this problem, this paper developed a recognition method of adhesive droplets based on Deeplab V3 + deep learning network, and a concave point matching segmentation algorithm based on convolutional neural network was designed to segment adhesion droplets. Used water instead of pesticides to spray to verify the detection effect of the above method. The test results showed that this method was accurate in the identification and extraction of droplets in the detection of droplet coverage. Compared with the droplet coverage measured by Deposit scan, the average relative error is 3.81%. Compared with the manual counting method, the detection error of the droplet deposition density was 3.17%. It was better for the segmentation of adhesion droplets. In terms of droplet size detection, compared with volume middle diameter measured by laser particle sizer, the detection error was 5.48%. This method improves the accuracy of UAV spray quality detection, so as to ensure the correctness of plant protection UAV spraying decisions.