Ephestia kuehniella eggs serve as an artificial feed for predatory mites, hence the industrial production of Ephestia kuehniella eggs is pivotal in breakthrough the bottleneck restricting the industrial cultivation of predatory mites. To solve the automation quality evaluation of Ephestia kuehniella eggs, machine vision technology based on deep learning is adopted in this study. Firstly a set of image acquisition platforms was established to acquire images, then the acquired images were annotated and amplified to get the dataset, then Faster R-CNN, a two-stage target detection algorithm, and a YOLOv5, one-stage target detection algorithm, were used to detect and count the data set. The experiment results demonstrate that Faster R-CNN achieved a detection accuracy of 0.93, significantly surpassing YOLOv5 of 0.68. Moreover, Faster R-CNN exhibited superior performance, particularly in scenarios involving overlapping eggs. This research lays a solid foundation for selecting an appropriate egg quality detection system algorithm for Ephestia kuehniella.