Non-destructive Detection Of Duck Eggs’ Surface Smudges And Freshness

The surface fouling degree of duck egg is much more serious than that of chicken eggs. More over, the more serious the smudginess is, the more microorganism there are. Then, the cross infection gets easier. At present, the detection of egg surface smudges still rests on the smudge detection, but not its classification according to the fouling degree. As to the freshness detection of duck egg, most of the egg quality detection still depends on experience detection manually. Because of the restriction of human vigor and reaction capacity, this not only may increase cost, but also have a great influence on detection result. Therefore, doing a research on the rapidly egg quality detection is of great scientific significance and realistic application prospect.In this research, duck eggs are the researching object, and machine vision technology is used to collect duck egg images, preprocess and extract characteristic parameters. Then, the fouling area ratio which is short for the ratio of duck egg’s surface fouling area to the whole area has been gotten, as well as the number of smudges. Also, in the duck egg’s transmission images with white shell, the yolk area ratio has been obtained as well as air chamber area ratio. Lastly, the related detection and classification model has been established which makes the duck egg’s surface smudges and freshness detection and classification come true. The following contents are the researching results.(1) The image acquisition system of duck egg’s surface smudges and freshness has been established. Several monocular egg candling lights were used to illuminate dirty duck eggs by the method of transmission. And every duck egg was photographed three times in order to ensure the integrality of collected images. As to rough cleaning duck eggs with white shells, single egg candling light was applied to illuminating eggs by the method of transmission. And the images of detecting objects were collected by employing high-definition camera convenient for the extracting of characteristic parameters.(2) The preprocessing method of duck egg images has been determined. The light-leaking section was gained through the B component binarization. Then, the binary image was recombined into three-dimensional image. And the recombined image was subtracted from the original color image which realized the remove of light leak because of the gap between equipment and duck eggs. As to the light-leaking in duck egg’s freshness images, threshold value was set when the R component binarized. Also, it was recombined and multiplied with original color image so as to eliminate the light leak.(3) The characteristic parameters of duck egg’s surface smudges and freshness have been extracted. The fouling area ratio and the number of smudges were chosen as characteristic parameters to distinguish duck egg’s fouling degree. And the range of characteristic parameters were determined by several experiments. A straight line detection method of hough transformation was used to extract air chamber area. Duck egg’s yolk area was obtained by doing related processing on G component. The yolk area ratio, air chamber area ratio and the gray average of R, G, I components were chosen as characteristic parameters to distinguish whether a duck egg was fresh or not.(4) The classification model of duck egg’s surface fouling degree and freshness have been established. The range of characteristic parameters to distinguish duck egg’s surface fouling degree was determined. Synthesizing the distinguishing method of duck egg’s three photographed images, the classification model of fouling degree was established by using MATLAB software and the accuracy was above 95%. The least square support vector machine was used to establish classification model of duck egg’s freshness. According to the proportion of 2 to 1, all of the samples were divided into training set and prediction set. The above 5 indexes were used as the characteristic parameters. The classification model established, the accuracy of training set and prediction set were 96.92% and 93.85% respectively.