Research On Key Problems Of Ice Cream Stick Quality Detection

A variety of ice cream bar defects are generated in the process of production,which generates undesirable influence for automatic production of ice cream and persons employ,such as split might scratch the mouth cavity of the user and contamination might menace the hygiene and food safety.Removing the ice cream bars with defects is very necessary before they are launched on the market for the above reasons.The defects might endanger healthy and safety of the user and these are the significant contents in the Quality Assurance System(QAS)on account of that ice cream bar is served as the ancillary product of foodstuff.Present methods are incapable of detecting defects at the appearance of ice cream bar effectively.Therefore,this paper takes split,contamination and burr as the research objects,and studies the methods of slight crack detection,slight contamination detection and supple burr detection,which are lack of effective information and cannot be resolved with existing digital picture processing technique.The primary researches and contribution are as follows.(1)A method of coarse-to-fine identification is proposed with a view to the problem of split extraction especially the shallow crack in the surface of ice cream bar,which contains abundant wood fiber textures.Firstly,the split texture edges are extracted preliminarily from the preprocessed ice cream bar surface via the edge detection operator,which is according to the grayscale difference between split and background.But this brings about a mass of noise meanwhile,which is wood fiber texture actually.Then the feature that the split always starts at the edge contour of ice cream bar is used.The alternative split texture edges are acquired after further screening towards the lines from the above step.Immediately following,the feature is used which is the great gray level fluctuations of split texture in the main lobe area coverage and sidelobe area coverage.In this paper,the extraction model is proposed for extracting main lobe gray feature and sidelobe gray feature from the above alternative lines.At last,the split texture edge lines are recognized via the texture main lobe and sidelobe feature.Thus,the split detection is accomplished.The experiment results show that the FDR is as low as 9.83 percent on the premise of without inspection,which decreases 5.53 percent at least comparing to other algorithms.(2)The geometric shape,gray-level distribution and spatial position of mineral linetexture are very similar and indistinguishable with split texture on the ice cream bar surface,so a detection method based on integral texture feature and internal texture feature is proposed.Firstly,there are some differences on integral gray-level,width,edge brightness variation tendency and average state.And five characteristic quantities of texture differentiation description are defined,which are gray-level variation tendency,width variation tendency,width average,steepness,depth average and bright side feature.The above five features are acquired from texture width,depth and edge height,but the edges of split and mineral line are too dim to accurately determined.So a model of variable scale gauss fitting is proposed for the texture reference information acquisition in this paper,which is based on that transverse profile grayscale distribution approximates to gauss curve.Specifically,resemblance and difference of transverse profile grayscale distribution and gauss curve are analyzed,and then the specific mathematical forms of fitting function are confirmed.The gauss fitting mathematical model is established,and the ideology of variable-scale fitting is proposed aiming at the nonuniform striped textures on ice cream bar surface.Thus,the detailed implementation program of straight line fitting enclosing is confirmed.Then,the target texture region is divided into several son regions,and the son regions data is fit via the established variable-scale gauss fitting model for acquiring the fitting parameters.At last,the five eigenvectors are obtained via merging gauss fitting parameters of the son regions.Then the distinguish of split and mineral line texture is studied according to the fluctuation characteristic of the internal cross section grayscale minimum value,and the concept and extraction method of ridge line are defined.The adjacent subareas positional relation between points and 3-direction number chain code rule are defined for describing the morphology of ridge line.And the extraction model of above chain code feature is defined for extracting ridge line texture fluctuation characteristic value.At last,the BP neural network classifier is established,and it is trained with the acquired five global features and one internal feature,which are used to identifying the split texture.The experiment results show that the average classification CDR of this method is 93.06 percent,which increases 14.82 percent at least comparing to the other detection algorithms and proves the validity of this method.(3)In this paper,a method based on Union-Find Sets and absolute neighborhood is proposed aiming at the problem of unfixed slight contamination shape,size and asymmetrical gray-level distribution.Firstly,there are the characteristics that the adjacent pixels gray values are relatively close in contamination regions and normal regions without defects,and the gray difference is obvious between contamination regions and surrounding neighborhood.The contamination regions and background regions are separated respectively through clustering the surface of the ice cream bar into small areas with similar grayscale using Union-Find Sets algorithm.Then alternative contamination regions are preliminarily selected from small regions which are segmented using Union-Find Sets according to that grayscale value of contamination regions is lower than surrounding neighborhood.As the interior of contamination is cotton-shaped and hollow,which makes a contamination region divided into several tiny regions after arithmetic processing,so the regions nearby in space position got from last step are merged.Then the area threshold is set according to prior knowledge,and the noise regions with small area are removed.And Low Grayscale Mean(LGM)is defined making use of cotton-shaped contamination interior.Also,the Absolute Neighborhood is defined regarding the connected son regions got from last step as fundamental unit.At last,the contamination regions are recognized precisely via the LGM numerical relationship of alternative regions and surrounding neighborhood.The experiment results show that this method could extract contamination regions preferably,which improves significantly comparing to the other detection algorithms effects.(4)The method based on the position relationship of supple burr and ice cream bar edge is proposed aiming at the problem of that the supple burr is too slight to detect and it is susceptible to ice cream bar edge transitional zone.These two edge detection models are combined to detect the external burr and middle burr.Firstly,the cursor is built to detect the burr which is continuous on edge.Then the characteristic of burr region gray level distribution is used to detect the burr which is discontinuous on edge.At last,the burr is detected according to the position relationship of the edges acquired via these two edge detection method.A method based on gray tail characteristic curve and parallel template is proposed aimimg at the detection of internal supple burr.The gray tail characteristic is extracted in line with the low gray value on image column of burr flaw,thus,the tail characteristic values of image are formed in sequence as characteristic curve.And then the parallel model is defined to extract characteristic through traversal from characteristic curve and noise suppression curve,thus,the detection of burr is accomplished.The experiment results show that this method could extract burr regions preferably,which is with certain practical value.Furthermore,the method of gallery building and the preprocessing of ice cream bar image are introduced in this paper meanwhile.