The Research Of Unclosed-glume Rice Seed On-line Doubl-sided Identification And Eliminate System Based On Machine Vision

Rice is one of the most important food crops in China.Compared to the traditional rice, hybrid rice has the advantage of strong resistance, high yield. Due to genetic traits and seed climatic environment interaction, the unclosed-glume often exists hybrid rice seed. Unclosed-glume is one of important appearance defects, the unclosed-glume rice seeds are easily affected by moisture, fungi and insect invasion, which in turn reducing vigor and germination rate.The cuurent processing methods can’t automatically sort unclosed-glume and normal rice seed. Machine vision technology is a quick, non-destructive detection method, it is possible to identify unclosed-glune rice seeds on line with double sided image analysis.This study aims to use machine vision technology to identify and eliminate unclosed-glume rice seeds on line. The main contents and conclusions are as follows:1. Design and build unclosed-glume rice seeds on-line detection and classification system based on machine vision. The system mainly comprises:a single-granulating unit, visual inspection unit, unclosed-glume rice seed removing device and control analysis software. Through contrasting characteristics of four current single-granulating devices, we try to apply vibration plate and linear vibrating conveyor, to achieve rice seed transported by the disorder to the pitch. Transparent plexiglass slide is used as visual inspection channel, and one optical fiber sensor controls two CCD cameras, which can obtain rice seed front and back images. In the process of eliminating unclosed-glume rice seed, the hollow cup digital servoe rotate to drive baffle and air blowing are contrast to discharge, the results show that air blowing has short response time, the sensing head of fiber optic sensor closes visual inspection channel bottom, the distance of two adjacent rice seeds which can be handle successfully is reduced.2. Hough line detection and line number are used synthetically to determine the normal and unclosed-glume rice seed. The algorithm for static normal rice seed and unclosed-glume rice seed, correct identification rates are 93.1%,90.4% respectively. When the average slip velocity is 0.46 m/s, the algorithm for front-side images of normal rice seeds and unclosed-glume rice seeds, correct identification rates are 89.7%,87.4% respectively; for back-side images of normal rice seeds and unclosed-glume rice seeds, correct identification rates are 88.6%, 87.5% respectively; for front-side and back-side images of normal rice seeds and unclosed-glume rice seeds, correct identification rates are 85%,84.6% respectively.3. The multithreading is used to process the front-side and back-side image for rice seed. Only using a single thread, the average time is 120ms for one rice seed with front-side and back-side.Using multithreading directly to process front-side and back-side images of one rice seed, the average time is 80ms.Using multithreading for processing the front-side and back-side images of one rice seed (removing the impact of the stopper plate), the average time is 60ms.4. Chunyou 84, Zheyou 12, Chiyou 65 with certain difference in appearance size are carried out the whole online test, respectively. The results show that:when the average slip velocity is 0.625 m/s and air pressure is 0.6 Mpa, normal and unclosed-glume rice seed correct recognition rate are both over 80%, after blowing, the processing accuracy of system above 70%, the processing speed is 4 grains per second under that circumstance.