Robust Detection of Cracked Eggs Using a Multi-Domain Training Method for Practical Egg Production
Published in Foods, 2025
Traditional defect egg detection algorithms based deep learning have poor generalization and accuracy degradation when egg varieties, origins, and detection environments change, which limits the application of machine vision-based defect egg detection algorithms in actual diversified and large-scale production scenarios. This paper studies a generalization detection algorithm of defective egg domains based on multi-domain training. The eggs of different varieties, cleaned or dirty, and different shooting scenes in Huanggang, Wuhan, Qingdao, and other places are taken as the target domain, and the domain-invariant features of the source domain are extracted. Improving the detection performance of the detection model adapting to the unknown data distribution in different target domains can better apply to diversified and large batch detection tasks in the actual production of eggs, and provide a basis for the intensive and intelligent development of egg production.link