Frei, U. K. | Search bioRxiv

bioRxiv2026

Single-kernel near-infrared spectroscopy enables haploid kernel sorting in field and sweet corn using high-oil haploid inducers across diverse donor-inducer combinations

Key messageA single-kernel near-infrared reflectance spectroscopy-based sorter can effectively identify haploid kernels for doubled haploid production in field and sweet corn backgrounds. Doubled haploid (DH) technology significantly shortens the breeding cycle for developing homozygous inbred lines in maize (Zea mays). Manual sorting of haploids from a larger bulk of hybrid kernels in an induction cross is a major bottleneck in DH development. Automated systems based on near-infrared (NIR) reflectance spectroscopy can be valuable tools for rapid haploid sorting, provided that sorting accuracy is sufficient for incorporation into the DH process. In this study, we evaluated the accuracy of a custom-built single-kernel NIR (skNIR) sorter for classifying haploid kernels from 12 high-oil haploid induction populations generated from two sweet corn and two field corn donors and four high-oil haploid inducers (HOHIs). We evaluated several general classification models that can be applied without population-specific recalibration or prior genotyping, including models that classified haploids based solely on predicted oil content, as well as multivariate methods that used all wavelengths of the NIR spectra. The highest classification accuracy was obtained using a general multivariate support vector machine (SVM) model. When combined with the two best-performing HOHIs, the general SVM model accurately sorted induction populations from two of the three donor backgrounds crossed with these inducers. Two oil-based methods showed less accurate classification than the multivariate SVM model, due to overlapping oil content distributions across the two kernel classes. Overall, this study demonstrates effective skNIR-based sorting of haploid kernels from diverse induction populations using a single general model. The practical deployment of this instrument in maize breeding programs is discussed.

plant biology↗

bioRxiv2022

Self-Supervised Maize Kernel Classification and Segmentation for Embryo Identification

Computer vision and deep learning (DL) techniques have succeeded in a wide range of diverse fields. Recently, these techniques have been successfully deployed in plant science applications to address food security, productivity, and environmental sustainability problems for a growing global population. However, training these DL models often necessitates the large-scale manual annotation of data which frequently becomes a tedious and time-and-resource-intensive process. Recent advances in self-supervised learning (SSL) methods have proven instrumental in overcoming these obstacles, using purely unlabeled datasets to pre-train DL models. Here, we implement the popular self-supervised contrastive learning methods of NNCLR (Nearest neighbor Contrastive Learning of visual Representations) and SimCLR (Simple framework for Contrastive Learning of visual Representations) for the classification of spatial orientation and segmentation of embryos of maize kernels. Maize kernels are imaged using a commercial high-throughput imaging system. This image data is often used in multiple downstream applications across both production and breeding applications, for instance, sorting for oil content based on segmenting and quantifying the scutellums size and for classifying haploid and diploid kernels. We show that in both classification and segmentation problems, SSL techniques outperform their purely supervised transfer learning-based counterparts and are significantly more annotation efficient. Additionally, we show that a single SSL pre-trained model can be efficiently finetuned for both classification and segmentation, indicating good transferability across multiple downstream applications. Segmentation models with SSL-pretrained backbones produce DICE similarity coefficients of 0.81, higher than the 0.78 and 0.73 of those with ImageNet-pretrained and randomly initialized backbones, respectively. We observe that finetuning classification and segmentation models on as little as 1% annotation produces competitive results. These results show SSL provides a meaningful step forward in data efficiency with agricultural deep learning and computer vision.

plant biology↗