| 摘要 |
The number of measurement signals relating to the properties of a sample are multiplied with a projection matrix derived from an optimized solution for a parameterization of a matrix created as a starting solution. Calibrating a measuring system comprising a number of sensors which generate signals from which a number (N) of measurement signals relating to the properties of a sample are generated, where the latter signals are used to classify the sample, comprises: (a) creating at least one NxM matrix S0 as a starting solution, where M is less than N; (b) searching for a parameterization P(p) of a NxM matrix, the parameterized matrix for each selected input parameter (p) having independent linear rows; (c) using P(p0), which corresponds to the transponate of S0, to optimize a solution for P(p) in relation to a target function f(P(p)), which depends on at least some of the smallest and the largest distances between two classes projected with P(p), so that an optimized solution P(pODP) is obtained, the optimization being carried out on the basis of measuring signals from a set of samples with a known classification; and (d) using the optimized solution to generate a projection matrix SODP, which is multiplied with N in order to provide an indication of the sample class. Independent claims are also included for the following: (a) a computer program used to carry out the above process; (b) a digital storage medium with this program; (c) a second process for classifying a sample, in which a primary vector A formed by the measuring signals N is multiplied by an NxM matrix SODP in order to form a secondary vector B with M elements for use during the indication, where M is less than N and SODP is generated using the above process; (d) a computer program used to carry out the second process; (e) a digital storage medium with the program for carrying out the second process; and (f) the apparatus used to carry out the second process. |
