Detection Of Clods And Stones From Impurified Potatoes Using Laser Backscattering Imaging

Potato is one of the four staples in the world.The detection and removal of foreign body mixed in harvested potatoes has not always been effectively.It is a key industrial problem of the development of potato industry with the implementation of the "Potato Staple Food Strategy" in China.This study aimed at distinguishing clods and stones from harvested potatoes.Firstly,color imaging technology was feasible to detect clods and stones from harvested potatoes,but the steps of feature extraction were tedious and the robustness of results was poor.Secondly,laser backscattering imaging(LBI)technology was studied to overcome the problems of color images.The imaging wavelength was optimized and the universality of the method used in different planting areas and potato varieties was studied.In addition,the conveyor trajectories were also studied for optimizing the detection system and offer opinions.Research contents and main conclusions of this study were listed as follows.1)Detecting clods and stones from potatoes based on color and shape of color imagesColor features based on wavelet transform and image blocks and shape features based on contour gaussian filter in color images were proposed respectively.The two-class classification and three-class classification of support vector machine(SVM)was used to distinguish clods and stones from potatoes.The influence of color feature parameters on the two-class classification result was analyzed.It was found that the results by color were better obviously when the blocks number was 4x4.The overall accuracy rates of two-class classification by color were 98.07%and 97.80%in 2016 and 2017,respectively.The results by shape were better when the distance threshold was 5,the gaussian filter template was 5,and the variance was 30.The overall accuracy rates of two-class classification by shape were 95.32%and 96.34%in 2016 and 2017 respectively,and the overall accuracy rates of two-class classification using color shape fusion features were 97.94%and 99.08%respectively.The color played an important role after principal component analysis(PCA)of color shape fusion features.The influence sequence of three features was color shape fusion features,color features and shape features.The recognition robustness of stones by color is poor.Two-class classification results by three kinds of features was better than three-class classification.Two-class classification was more suitable for detecting clods and stones from potatoes.2)Detection of clods and stones from potatoes based on LBI technologyTo overcome the difficulties of the large amount of calculation and low robustness of color images,LBI technology was put forward to detect clods and stones from potatoes according to the scattering differences because of differences in organizational structures.On the preliminary experiment research,the features of scattering profiles referred to pulse signal parameters were extracted,and two-class mahalanobis distance discrimination(MDD)and three-class MDD were used to to classify clods,stones and potatos.In two-class classification results,5 kinds of accuracy rates were more than 99%.It was indicated that LBI technology was feasible to classify foreign bodies from potatoes.The results of LBI in 2017 was as good as that of color imaging technology,and LBI technology could be used to detect potatoes on-line.In order to optimize the wavelengths of LBI system,scattering line width and the fitting coefficients of lorentz function and exponential function as features were presented respectively.On the one hand,the classification based on scattering line width was directly according to threshold value;on the other hand,a probabilistic classifier was designed based on the recognition results of fitting coefficients of all profiles extracted for every sample,and the final classification was based on probability threshold.The receiver operating characteristic curves(ROC)and area under the curves(AUC)were used to evaluate the classification result.The wavelengths 780,830,850nm were selected with all kinds of accuracy rate over 98%by the two method,and the accuracy rates of clods and stones were both above 99%.In order to ensure that LBI technology proposed was adaptive to growing places and varieties of potatoes,5 kinds of filter feature selections were used to reduce feature dimensions for 4 groups of samples from different production places and varieties.Borda count was used to grade every feature based on 5 sequences.The feature numbers were determined according to overall accuracy rate with orderly increasing the feature numbers in the whole rank of each sample group.The common features of the 4 groups were selected as the results of feature selection when the overall accuracy rates were stable,which were coincident with the average discriminant results of SVM,MDD and linear discriminant analysis(LDA)based on a single feature.There were 8 features selected.It was found that the recognition result of the mixed model of 4 groups was influenced by impurity rate,and the results were poor when the impurity rates were lower.Classification results of three classifiers of the prediction sets were over 90%at different impurity rates when impurity rates of training sets were fixed at 40%,50%and 60%.The 5 kinds of accuracy rates of prediction sets were almost all over 97%by using SVM under different impurity rates when the impurity rate of training set was 60%.This results basically met the actual production and the method could be used in online detection syetem.It was proved that the LBI technology was feasible to be applied on-line for detecting the foreign body of harvested potato.3)The key parameters of the potato conveyor trajectoriesThe influence of initial postures on the conveyor trajectories was studied in simulation experiment andactual experiment.It was found that the potato trajectories were influenced significantly by initial postures.Further,the influence extent was related to initial speeds.The influence was decreasing with initial speeds increasing,and there would be a stable state as initial speed higher than 2.0 m/s.The reason of the same potato with different trajectories was rotation occurrence in falling.The direction of camera could not be decided because of that under different initial postures and initial speeds.Calculation methods of guide mechanism length and installation location of the unloading conveyor were presented.The guide mechanism width was 200 mm at falling height 200 mm,and the x-coordinate of unloading conveyor at falling height 400 mm was 174 mm when the initial orientation influence was ignored at initial speed 1.0 m/s.