Plant Breeding Evaluation Based On Coupled Feature Representation

With the rapid development of improved breeding equipment and information technology, computer-aided decision-making in plant breeding evaluation can help solve the problems associated with high-throughput demand and insufficient experience of breeders in modern large-scale field breeding experiments. Many linear models have made great contributions to the development of breeding evaluation although they are based on a wrong assumption of attribute independence. This paper proposes a unified coupled representation that integrates intra-coupled and inter-coupled relationships to capture the interdependence among quantitative traits by addressing coupling context and coupling weights. Moreover, a hybrid scheme of the linear correlation and ordinal relation is introduced to express the coupling relationship with a preset parameter that balances the contributions so as to capture both relative and absolute performance in cultivar selection and breeding evaluation. A framework that includes data preprocessing, coupled data representation, feature selection, prediction model construction, and assisted decision-making is our overall solution for the plant breeding evaluation task. Experiments on real plant breeding data sets demonstrated the effectiveness of coupled representation for elucidating the quantitative phenotypic traits and the advantages of the proposed plant breeding evaluation algorithm compared with benchmark algorithms.