In order to research the problem of unfertilized eggs verification before hatching, combined the technology of digital image processing and embedded technology, a design scheme of unfertilized eggs filtering system was put forward based on embedded computer. At first, the images of eggs that were collected by using computer vision technology were applied, and then the invariance parameters of image area were extracted, such as circularity, complexity, elongation, globularity, ratio of major to minor axis and average alteration coefficient. At last, identification of unfertilized eggs based on trained genetic algorithm neural network was realized. The test result showed that the system was convenient and easy to realize. Using 600 non-samples of the hatching eggs as verification samples and the result showed the identification rate could reach to 99.3%, satisfied the requirement of eliminating the unfertilized eggs before hatching.