Extraction Of Key Quality Indexes Of Rice Based On Image And Spectrum Analysis

Improving the quality of rice is of great significance to meet people’s growing needs for a better life,among which rice quality inspection is the basis.The traditional inspection method has low efficiency,high cost and low precision.In this paper,key quality indexes of rice are extracted based on image and spectrum,which is nondestructive,efficient,convenient and accurate.The main research contents include as followings:1.The image acquisition method and complete image preprocessing process are designed.A LED flat lamp is used as transmission light source and an ordinary SLR camera is used to collect images,which are very easy.By acquiring the camera’s internal and external parameters,undistortion,ortho-rectification,removing the uneven background and other preprocessing,the geometry accuracy of image and the applicability of the algorithm under various imaging conditions are guaranteed.2.A complete process of broken rice recognition and touching rice split is designed.This paper proposes a method of getting length threshold adaptively to distinguish broken rice from head rice,which is based on the concentration of length distribution of head rice and does not need training samples.For touching rice kernels,after identifying by concavity,the touching points are detected by curvature analysis of the contour,and the angle weighted minimum distance method is proposed to pairing the touching points to form a split line,which is better than the traditional minimum distance method in the case of complex touching.Through the experiment of 32 kinds of rice,as a whole,the precision rate,recall rate and F1 value of touching rice split are98.70%,98.99% and 98.84% respectively,and the relative errors of counting of all rice,broken rice and head rice kernels are-0.13%,0.11% and-0.18%,respectively.3.Inspection of appearance quality was studied.For inspection of shape,the precision of the length,width and aspect ratio of 5 kinds of rice by the longest distance method,the polar coordinate method and the minimum enclosing rectangle method are compared.It is found that the precision of the minimum enclosing rectangle method is the highest,and the error is less than 0.1mm.For inspection of chalkiness,the process of chalkiness segmentation is improved and chalkiness is classified.In this paper,chalkiness is enhanced by gray-scale transformation,and the most likely chalky rice is selected by chalkiness index and used to get gray distribution histogram,then the chalkiness is segmented by the combination of two thresholding methods.This method is more applicable than the traditional method of minimum intra-class variance.The difference of the chalky rice rate and chalkiness degree between the artificial inspection and the method in this paper is less than 5% and about 1% respectively among 12 kinds of rice.The accuracy of embryo position recognition of the adaptive segmentation method proposed in this paper is 96.9%,which is higher than that of the longest polar method and Harris corner detection method and needs no training sample.According to the position of embryo,chalkiness can be divided into white belly,white center and white back.4.The extraction of amylose content based on reflection spectrum was studied.In this paper,the partial least square regression is used as the modeling method,and it is found that the multiple scattering correction can improve the accuracy,and the model based on sensitivity bands selected by the successive projections algorithm is better than the model based on full spectrum and the principal component.The root mean square error of cross validation of the optimal model is 1.08,and the determining coefficient is 0.63.The research results of this paper can realize the automatic extraction of key quality indexes of rice,which has certain reference value for the algorithm design and equipment development of rice quality automatic inspection.