A System For Counting Overlapped Quasi-circular Granule By Machine Vision Technology

In the industrial and agricultural production,We usually need to count the number of some kinds of circular particle,for example we determine grain weight of food and edible oil.Manual counting method is the most primitive method,and the long time in statistical process crops would cause extreme eye fatigue.As a result,the accuracy and counting speed are greatly affected by human factors,especially on the occasions of high requirement for efficiency and accuracy,its limitation has greatly affected the efficiency of production testing.In order to improve agricultural crops statistical accuracy of particles and labor productivity,we urgently need to develop a system which can automatically measure the number of crops.In this paper,a pile of crop particles are counted after taking its photographs and extracting from the target particles.An effective identification and intelligent segmentation algorithm are accounted for improving the accuracy of counting circular particles which are profoundly overlapped and adhered.Firstly we collect the products grain color image,after be preprocessed and firstly partitioned,the region of all overlapped grain is distilled;The concave points and grains' approximate center of area are got by using hollow-point search method and Hough transform.The Euler number of topology shape is found out in the paper.And then establish the characteristics database of particles overlapped categories.The particle superposition types are intellectively transformed into serial,parallel and double-deck overlapping types by using Support Vector Machine training method.This method automatically divides overlapped granule into lots of single grains by structuring the corresponding off-line after the process of matching these hollow points,Finally the number of particles are count out.Through random trials of soybeans,mung beans and other crops,the system could achieve the result rapidly.Accurate segmentation rate is up to 99.8 percent,and the counting error rate is less than 0.5%after separated the overlapped particles.The experiment results show that this method is high precision,efficiency and the ability of anti-interference when it is used in counting the particles.A method for segmenting and counting overlapped granule is proposed in the foundation of the machine vision technology and image processing technology.We will make a comprehensive research the support vector machine classification algorithm which provides a theoretical support for partition.This counting system which selects DM643 as the main processor has a wide range of applications,such as in space of measuring the weight of 1000 crops,identifying and taking the count of medicine biological cells.