Maize Purity Identification Based On Machine Vision And Improved DBSCAN Algorithm

Corn is one of the main crops in China and is used widely in food industry and feedstuffindustry while its purity influences greatly the quality and production. Nowadays, the mainmethod of corn purity identification are biochemical electrophoresis, chromatographicanalysis, fluorescent test and DAN molecular markers techniques which have high accuracybut too long frivolous cycle for using in rapid batch examination. Achieving automated andnon-destructive corn purity identification with machine vision is significance to enhance thelevel of seed purity identification technology and promote seed industry healthy progress. Aswell in this paper corn purity identification with machine vision is studied. The main contentsof the paper are as follows:(1) Detection system of corn purity identification based on machine vision is built. Ablack box as surroundings is designed to improve the image quality. After experimenting，shadowless lamp is used as the light source and black is chosen as background based on itshigh contrast with corn color.(2) The original images are pretreated based on the feature extrication of corn crown areaimage. Firstly, color image gray processing is made. Compared to histogram equalization andmean filter, median filter could achieve better effect on image enhancement. Thresholdsegmentation and region growing method are used for image segmentation to separate thetarget images from background as well as the single target image is separated.(3) Base on the difference in shape of maize grain，the corn crown area is chosen as theresearch object. Study the area to make sure the core area. Nine color features parameters areextract On the basis of RGB，HIS and Lab three color modes. In order to reduce thedimension of vectors and improve the identification rapid, these9eigenvector are analyzedand optimized. Taking the Using the mathematical expectation，mean square and variationcoefficient to calculate statistical regularities of this characteristic parameters, variationcoefficient illustrates the level of difference of mathematical expectation while the standarddeviation proves the degree of dispersion about characteristic value， the discrete levelbetween hybridization corns and others. Confirmatory analysis showed that H, S and B areextracted as feature vector and density contrast in theirs3-D localities. (4) Maize purity identification based on improved DBSCAN algorithm by farthest firsttraversal algorithm (FFT) is presented for the density contrast in (H, S, B)3-D localities.Firstly, the data is improved by FFT to strip abnormality points of the side out. Secondly,DBSCAN can handle spatial high density data by dividing them into clusters with highenough density. Finally, the purity identification findings are output by combining the aboveresults.(5) With four common samples of Nongda108， Zhengdan958， Jundan20andLudan981for this study, these methods and algorithm models are used to distinguish the cornpurity base on the color characters of crown core area. Simulation results show theeffectiveness of the proposed technique by accurate identification reaching to93.3%whichyields better performance than traditional approaches.