The Detection of Eggshell Cracks Using Different Classifiers

Abstract

Chicken eggs, which are widely consumed in daily life due to their rich nutritional values, are also used in many products. The increasing need for eggs must be met quickly for various circumstances. Eggs are subjected to various impacts and shaken from production to packaging. In some cases, these effects cause an eggshell to crack. While these cracks are sometimes visible, they are sometimes micro-sized and cannot be seen. The cracks on the egg allow harmful micro-organisms to spoil the egg in a short time. In this study, acoustic signals generated by a mechanical effect to the eggs were recorded for 0.2 seconds at 50 kHz sampling frequency using a microphone. To determine the active part in the collected acoustic signal data, a clipping process was implemented by a thresholding process. Thus, the exactly correct moment of mechanical contact on the eggshell was easily detected. After passing the determined threshold value, statistical parameters such as min, max, difference, mean, standard deviation, skewness and kurtosis were extracted from the data obtained, and 7-dimensional feature vectors were created. Finally, the Common Vector Approach (CVA) is applied on the extracted feature vectors, 100% success rate has been achieved for the test data set. The ANN and SVM classifiers in where the same feature vectors are treated were used for the comparison purpose, and exactly the same classification rates are attained; however, the less number of eggs are tested with the ANN and SVM classifiers in the same amount of time. With the proposed mechanical system and classification methodology, it takes about 0.2008 seconds to determine whether the shells of eggs are cracked/intact. Therefore, the proposed combination of the feature vectors based on statistical features and CVA as a classifier for the detection of cracks on eggshells is notably appropriate especially for industrial applications in terms of speed and accuracy aspects.