This paper studies the parameter estimation and classification method of matrix data under the matrix normal dis-tribution.Firstly,based on the low-rank decomposition and the penalized likelihood function method of matrix normal distribution,a method for parameter estimation and adaptive determination of rank of matrix data is proposed.Then the block coordinate de-scent method and the augmented Lagrange multiplier algorithm are used to give an effective iterative estimation algorithm.Fur-thermore,based on the discriminant analysis method,the rule of classi fication and prediction under low-rank decomposition is proposed.Finally,through the application of a large number of numerical simulations and the recognition of satellite land resource data and handwritten digits,the low-rank estimation method is proved to be effective in improving the estimation and classification prediction accuracy of matrix data.
matrix normal distributionlow-rank decompositiondiscriminant analysisclassification prediction
国家科技期刊平台
NSTL
万方数据
