Classification Of Diospyros Lotus Seeds Based On Machine Vision

The rapid classification of germplasm resources is of great significance to the development of national germplasm resources.With the development of information technology,image processing and recognition technology play an increasingly important role in seed classification detection.Based on the computer image processing technology,this paper comprehensively uses the knowledge of graphic image and pattern recognition,The automatic identification and classification of the seeds of diospyros lotus were studied.This paper mainly includes the following research contents and conclusions:1.The machine vision system of diospyros lotus seed image acquisition was built.The system is mainly composed of light source,camera and computer.The system uses a programmable industrial camera to achieve automatic collection of diospyros lotus seed images.2.Image preprocessing.By analyzing and comparing the results of several image preprocessing methods,including image gray scale processing,image filtering,threshold segmentation,morphological processing,the target seed is extracted,and then detect the edge of the seed.The algorithm is preprocessed by a variety of algorithmic operators.Analyzed and compared all the results,and the optimal algorithm is selected.3.Seed feature extraction.This paper analyzes the seed shape and color characteristics of diospyros lotus seed,including 13 geometric features: area A,circumference P,circularity e,long axis L,short axis S,aspect ratio w,maximum radius Rmax,minimum radius Rmin,radius ratio z,long axis 1/4,3/4,1/8,7/8 length and 6 color characteristics: the mean and variance of R,G,B colors.Image processing technology is used in feature extraction algorithm,and HALCON software programming to achieve the characteristics of automatic extraction and storage.In the extraction of seed characteristics,a seed tip recognition method based on machine vision was proposed.The seed tip position was determined by segmenting and fitting the seed contour.The results showed that the comprehensive accuracy of seed tip recognition was 83.6%.According to the seed tip position,the seed spindle direction is determined.Then,the calculation method of the distance between any two points of the seed contour is given.The measurement method is not affected by the position of the seed and the front and back,which can distinguish the length of the minor axis of the seed and calculate the results quickly.4.Modeling and Identification of Seed Varieties.The neural network algorithm was used to establish the model of five kinds of seeds,and the characteristic parameters of 19 were input into the neural network.By data analysis and experimental verification The accuracy of the method was 91.8%.Based on the analysis and comprehensive evaluation of the whole classification system,the development direction and suggestion of seed classification based on machine vision technology are put forward.