Study On Detection Methods For External Features And Crack Of Duck Eggs Based On Machine Vision

Egg industry is one of the pillars of agricultural economy in China, yet it is not competitive in international markets. One of the reasons for this phenomenon is due to the fact that China lacks advanced egg quality detection methods. Duck eggs are usually littered with dirt, weeds and feather due to the habitat of duck which is surrounded by water. It is difficult to detect the quality of duck eggs for complex smudges on the surface. Currently, the quality detection of duck eggs is conducted manually to a great extent. The quality of duck eggs cannot be guaranteed. Therefore, the study on intelligent detection methods for duck eggs is of great significance in both academics and industrial applications. In this study, machine vision was used to detect the external features and crack of dirty duck eggs. The contents and results were shown as follows.(1) The image acquisition system was set to capture the images of external features and crack of duck eggs. Monocular lights were used to illuminate duck eggs by transmission. Every duck egg was photographed three times with different rotations to image the whole surface of the duck egg.(2) The image preprocessing method was determined. The undesirable effect of light leak caused by the space between the devices in the images of external features and crack of duck eggs was minimized by deducting component B from component R and image masking, respectively.(3) The ellipse fitting algorithm of duck egg’s edge was determined. Least squares algorithm and least median of squares algorithm was used to ellipse fitting the edge of duck eggs, respectively. Through comparative analysis, least median of squares algorithm was determined to fit the ellipse.(4) The characteristic parameters of duck egg’s external features were extracted. The values of the semi-major axis, the semi-minor axis and the volume of fitted ellipse were extracted as characteristic parameters of duck egg’s size; the values of the semi-major axis, the semi-minor axis and the eccentricity of fitted ellipse were extracted as characteristic parameters of duck egg’s shape.(5) The grading model of duck egg’s external features was established. Neural network and support vector machine was used to establish the grading model of duck egg’s external features, respectively. According to size, duck eggs could be graded into small, medium and big; the duck eggs could be further graded according to shape into flat, medium and round. Through comparative analysis, support vector machine was determined to establish the grading model of duck egg’s external features. The correct grading rate of size and shape was 93.33% and 95.00%, respectively.(6) The feature extraction method for cracked duck eggs was determined. Various edge detectors and texture analysis functions were used to extract the feature of crack. Through comparative analysis, local range value was determined to extract the feature of cracked duck eggs.(7) The detection method for cracked duck eggs was determined. All three images of each duck egg would be detected. If there was no crack in three images, the duck egg was intact, otherwise the duck egg was cracked. The recognition rate of intact duck eggs and cracked duck eggs was 86.67% and 96.67%, respectively.