A feature extraction and identification method of corn haploid and hybrid kernels was studied. The corn kernels were produced by hybrid induction after importing gene markers to their male parent in corn haploid breeding process. The proposed method made the accurate identification of corn haploid kernels come true. Corn kernels produced by hybrid induction was researched. Firstly, the corn kernels were divided into three parts according to the color distribution, including purple marks, yellow endosperm and white embryo. Then a (G-B)R/B model was proposed by analyzing color features of pixel samples that got from the three parts. This model could extract the purple marks area effectively. A simple and quick identification method of corn haploid kernels was proposed after researching on morphological characteristics of corn kernels and distribution characteristics of purple marks. Experiments finished in the corn haploid sorting platform, they showed the accurate rate of corn haploid identification was above 98.07% by the proposed method, and it met the requirements of dynamic test.