A Comprehensive Review of CNN Based Image Classification for Egg Fertility Detection

The integration of technology into agriculture, particularly in poultry farming, has transformed processes like egg fertility detection, which is critical to improving both productivity and economic viability. This paper reviews the evolution of Convolutional Neural Networks (CNNs) as a non-invasive solution for egg fertility detection, focusing on their growing application in automating this traditionally labour-intensive task. Conventional methods, such as candling and chemical testing, are discussed, with attention to their historical importance and inherent limitations in accuracy and scalability. Early efforts in automating egg fertility detection, utilizing techniques like ultrasound, electrical conductivity, and visible transmission spectroscopy, are analysed to highlight their advancements and ongoing challenges. CNNs have revolutionized image classification and, in recent years, have shown significant promise in egg fertility detection. The paper covers the architecture and function of CNNs, detailing models such as LeNet-5, Mask R-CNN, YOLO, Transfer learning the latest DPSA network. Each of these models is evaluated for its accuracy, efficiency, and ability to differentiate between fertile and non-fertile eggs. A comparative analysis demonstrates the superiority of CNN-based approaches over traditional methods, emphasizing improvements in detection speed, reliability, and the potential for large-scale application in poultry farming. The review concludes by exploring future research directions, focusing on the potential of CNNs to further enhance non-destructive fertility detection methods. The need for continuous innovation in automated systems is underscored, particularly as the global demand for poultry products grows and as inefficiencies in current detection methods, including the economic impact of unfertilized eggs, drive the need for more accurate, scalable solutions.