基于稀疏低秩分解的杂草种子配准

Weed seeds registration based on sparse and low-rank decomposition

针对杂草种子识别在实际应用中的困难,提出了一种适用于杂草种子配准的稀疏低秩分解算法。阐述了稀疏低秩算法的原理和求解方法,原本有等式约束且非凸的问题可以通过求解核范式和l1范式的无约束凸优化问题得到很好的配准结果。为了验证配准工作的重要性,运用k折交叉检验对比配准前后的识别率差异。实验结果表明,基于稀疏低秩分解的配准算法能够提高杂草种子的识别率,为实际中的杂草种子识别提供了可行方案。

It is difficult to apply the weed seeds recognition method in real world. A sparse low-rank decomposition algorithm is proposed and introduced for weed seeds registration. The problem originally having equality constraints and being non-convex is obtained by solving the nuclear norm and l1 minimization problem with unconstrained condition. To verify the importance of registration, k-fold cross validation is used to compare the results. Experimental results indicate that the recognition rate based on this registration algorithm can improve the recognition rate. A practicable solution is established in the future.