Study On The Performance Detecting Technique Of Seed-metering Device Based On Machine Vision

Detecting accurately the technical parameters of seed metering device is the key to test the device performance and develop precise seed-metering device.This paper analyses the insufficiencies of butter tape test bench, suggesting developing seed-metering test bench with sand-paving conveyor belt and preventing the seeds from bouncing through the buffer action of sand layer. Machine vision technology is also applied to collect and process the seeding image to realize the automatic recognition of seed and seed spacing measurement, which, on this basis, achieves the real-time automatic analysis of the device performance parameters.This paper mainly discusses the following researches and draws the relevant conclusions:(1) Seed-metering test bench is developed by applying the principles of paving sand. Compared with traditional way of oiling, this bench owns advantages of seed recyclability, non-pollution, and background with good machine vision and so on.(2) Photography system is constructed by applying the industrial cameras and light. A stable photograph effect can be guaranteed by setting special photograph studio. The image is calibrated and camera parameter is optimized.(3) The control system based on RS 485 Bus is established, which realizes the precise control of all moving parts.(4) Sequence seamless image mosaic technology is put forward innovatively based on the combination of induction reference mark and image reference mark, in which induction reference mark ensures the stability of image sequence under different speeds of conveyor belt and image reference mark realizes the seamless mosaic of adjacent images.(5) For the image features, corresponding image processing procedure is formulated which mainly includes image region segmentation, image segmentation, connected region labeling, image feature amount extraction and image seamless mosaic.(6) Image processing algorithms is analyzed and optimized, for example, different threshold calculation methods are used for different image recognition features; identification of reference mark and seed is realized based on the target recognition algorithm of connected region labeling; virtual image mosaic method is proposed by saving image sequence location data.(7) Qualitative test analysis is conducted for the image acquisition and mosaic effect which turns out that mosaic image and the actual image has a good correlation and proves the feasibility of image processing algorithms.(8) Qualitative and quantitative analysis of seed identification and location and measurement of seed spacing are conducted for the image which shows that the results of image processing are in line with the actual distribution of seeds.(9) The system is applied to the seed-metering device to test the performance of the metering accuracy which, compared with the number of manual measurements, turns out that system performance meets the requirement of detection.