Maize Haploid Identification Using Hyperspectral Imaging and Dual-Branch CNN

Haploid breeding technology is an important technology in modern crop improvement and genetic engineering, which can shorten maize breeding periods and enhance breeding efficiency. Haploid identification is a key process in this breeding technology. In this paper, a hyperspectral imaging system with a spectral range of 866.4-1701.0 nm was used for nondestructive identification of maize haploid seeds. To improve the accuracy of identifying haploids, a dual-branch convolutional neural network (DB-CNN) fusing spatial-spectral information is proposed. The experimental results show that DBCNN obviously is superior to several existing methods and obtains 99.00% overall accuracy. Compared with the traditional methods, it can fully extract the effective information in the spatial and spectral domains and has a stronger discriminative ability.