An Automated Candling System for Duck Egg Fertility Detection, Sorting, and Counting via Digital Image Processing

Abstract

Duck farming is considered one of the lucrative livelihoods in the Philippines because of its many advantages with minimal cost and high-profit returns. This is the second-largest poultry industry in the country, yet this industry is still immature and underdeveloped. There were many attempts done to improve the egg production sector of the poultry industry, particularly in the candling method. However, they mainly focused on chicken eggs. Hence, automating the candling system for the incubated duck eggs would be beneficial. The main objective of this study was to create an automated candling system for duck egg fertility detection, sorting, and counting via digital image processing. The hardware of the proposed system was made from locally available components, materials, and equipment. The most suitable light source was assessed, and it was found out that a 9W White LED would give the optimum results among the tested light bulbs. The camera was also calibrated and was placed three inches away from the incubated duck egg and was mounted ninety degrees from the horizontal. The fastest allocated processing time that would give a high fertility detection accuracy was assessed and was determined to be 2.5 seconds. The classification model was created in TensorFlow and was evaluated using a confusion matrix. The results showed that the classification accuracy was 0.86, f-measure was 0.875 and MCC was 0.7417. The functionality was also assessed and was 100% functional. The accuracy of the machine to detect the fertility of 5-, 6-, 7-, 8-, 9-and 10-day old, incubated duck eggs were also tested and were found out to be 50%, 60%, 90%, 100%, 80%, and 80% accurate, respectively. The counting of total inputted duck eggs and total sorted "balut" and "penoy" was also 100% accurate. Lastly, the efficiency of the system was evaluated by calculating the overall equipment effectiveness and was found out to be 83.99% efficient.