The use of a matrix decompositions for dimension reduction of training sample

Proposed two-step procedure of dimension reduction training sample for pattern recognition neural networks deep learning. The use of the procedure improves the efficiency of object recognition in control systems due to the substantial reduction of computation.