Principal component analysis-based three-dimensional quantitative characterization of cracks on perforated embryo eggs.

During vaccine production using the chicken embryo cultivation method, viruses such as influenza need to be inoculated into 10-day-old chicken embryos for cultivation. Prior to virus injection, perforation on the eggshell may induce cracks, leading to potential cultivation failure. This necessitates identifying and removing unqualified cracked perforated embryo eggs. Traditional two-dimensional methods for egg eligibility judgment, limited by precision and analytical dimensionality, only consider crack presence while ignoring crack acceptability, resulting in erroneous rejection or missing detection. This study proposed a global-local three-dimensional(3D) quantitative characterization framework based on principal component analysis(PCA). Through image processing and 3D reconstruction, global and local point clouds of the perforation-induced cracks were generated, fully preserving their spatial information. PCA was employed to perform global quantitative characterization of the crack geometric information, ultimately yielding 7 global feature parameters. A crack shape identification model was constructed based on the random forest algorithm, achieving a mean accuracy of 99.19%(standard deviation 0.87%) and a macro F1-Score of 98.91%(standard deviation 1.19%). For cracks requiring further evaluation, local quantitative characterization was conducted to extract perforative cracks' local geometric features parameterically. This framework achieved precise 3D reconstruction and quantitative characterization of perforative cracks, providing quantitative parameters that could support crack acceptability evaluation. It has the potential to promote perforated egg eligibility judgment from 'crack presence' to 'crack acceptability,' thereby facilitating potential automated screening of perforated embryo eggs under different virus cultivation conditions with varying crack acceptability criteria.