In industrial agricultural breeding, double haploid based generation of inbred maize lines has accelerated the time to market of commercial seed varieties [5]. Traditionally, haploid corn seeds are manually discriminated from the diploid seeds using visual indications of the molecular marker system that is selectively expressed in the embryo region of the diploid seeds. In the industrial scale, there have been two notable automation efforts based on the R1-nj marker system [2, 4]. However due to the extensive phenotypic variation of the marker expression [1] and heterogeneity arising from image acquisition in the field, developing computer vision methods to classify seed images is challenging, and approaches robust in recovering haploids are lacking.
展开▼