Study On The Key Technology Of Peanut Quality Sorting System

Peanut is an important economic crop and oil crop in China.It has high application value and broad development prospects in the fields of agriculture and industry.The size,shape,mildew and damage of peanut are clearly stipulated in the export of foreign trade,but current sorting machine can only sort the mildew grain,which is difficult to meet market demand.Machine vision technology is increasingly being applied to the quality identification of food or crops,especially the research and implementation of machine vision technology and machine learning algorithms developed in recent years.Using machine vision technology to realize automatic nondestructive testing of peanut quality has important theoretical value and practical significance for improving the market competitiveness of peanut and the healthy and sustainable development of peanut industry in China.In this paper,the key technologies of the peanut quality grade sorting system are studied,including the analysis,research and comparison of various feature extraction methods of peanuts,as well as the research,design and analysis of sorting algorithms.The main research contents and conclusions are reflected in the following aspects:(1)The preprocessing method of peanut images is studied.The 3×3 median filter is selected for filtering and denoising through experiments.By analyzing the values of three channels in the foreground and background of the peanut image,the linear combination of RGB is used to segment the peanut image,so as to reduce the influence of the brightness of the light source on the segmentation effect,and at the same time to avoid the disadvantage of Ostu,which has higher operation efficiency.(2)The extraction methods of texture features and color features of peanut images are studied.The texture analysis region is cut off and the texture features are extracted by Gauss-Markov Random Field.Based on HSV color space,histograms and color moments are used to extract color features respectively.(3)The contour feature extraction method of peanut is proposed.First,the ellipse is obtained by Direct Least Squares Fitting,and the fullness and symmetry feature extraction method are proposed respectively,and then compared with those of other researchers in terms of feature parameter distribution and correlation coefficient.The experimental results show that the fullness feature extraction method proposed in this paper can more effectively measure the fullness and standardization of peanut shape,and the correlation coefficient is increased by 0.129 or more compared with other methods.The distribution of the two symmetry features on the two kinds of samples is quite different.Moreover,these three characteristic parameters can be calculated synchronously and thus have high computational efficiency.(4)A sorting algorithm with higher computational efficiency is proposed.The algorithm generally has a branch judgment structure,and each classifier inputs different types of features according to different discriminative targets.Then the design and training methods of these classifiers are introduced.Neural network is used as classifier A,and its cost function is weighted appropriately according to the characteristics of dataset during the training process;Linear discriminant analyzer is adopted as classifier B;Linear support vector machine is used as classifier C,the feature selection method based on the classification correct rate feedback is introduced,and the relationship between accuracy and the number of features is explored through cross validation.(5)According to the sorting algorithm adopted by other researchers in recent years,a comparative experiment is designed and compared with the algorithm in this paper in terms of accuracy and operational efficiency.The experimental results show that compared with the contrast experiment,the overall accuracy of the proposed algorithm is improved by 3.34 percentage points,the Kappa coefficient is increased by 0.0436,which reflect an improvement in accuracy.The average number of extracted features decreased from 28 to 16,and the amount of calculation decreased from 396 to 76,which reflect a significant increase in operational efficiency.