Method and apparatus for classifying known specimens and media using spectral properties and identifying unknown specimens and media

Apparatus for performing a computer-implemented method of determining a similarity metric for use in classifying an unknown specimen to a reference group of a plurality of different reference groups of a reference collection of spectral data for reference specimens in a spectral database and predicting a value of a trait of the unknown specimen, the apparatus comprising:

• a data processor for reducing a dimension of input spectral data of the spectral database;
• for receiving input for selecting a similarity metric of a plurality of different similarity metrics to classify the unknown specimen to the reference group of specimens;
• determining a threshold responsive to the selected metric for classifying the unknown specimen to the reference group of specimens;
• and predicting the value of a trait of the unknown specimen by determining membership of the unknown specimen in the reference group of specimens the spectral database storing a value of a trait of the reference group of electrical device specimens in the spectral database memory for output once membership of the unknown specimen in the reference group is determined by the data processor according to a similarity threshold;
• and
• a spectrum analyzer for collecting spectral data over a frequency range for input to the spectral database of the spectral database memory of the data processor, the input spectral data comprising a magnitude for each frequency for which spectral data are collected.

A computer implemented method for classifying an unknown specimen having spectral data to a reference group of a plurality of different reference groups of a reference collection of specimens of a reference spectral database of a data processor, a reference specimen having a first trait, the method comprising:

• storing measured spectral data for a reference group of a plurality of different groups of reference specimens in said reference spectral database of data processor memory, each reference spectral data specimen of the spectral database comprising spectral data of magnitude at a frequency and a value for the first trait associated with the reference group of the spectral database;
• generating, an index for the spectral database having data objects, each data object comprising a vector of attributes the attributes comprising one of real and imaginary parts, frequency, magnitude and base phase angle and equivalent means for defining a spectral vector;
• determining a similarity threshold for membership in the reference group of reference specimens using a selected metric of a plurality of different similarity metrics,
• classifying the unknown specimen having a different trait as belonging to the reference group of the reference spectral database using the selected similarity metric of the plurality of different similarity metrics;
• and associating the different trait of the unknown specimen as belonging to the group of reference specimens having the first trait.

A computer-implemented method of determining a similarity metric for use in classifying an unknown specimen having spectral data to a reference group of a plurality of different reference groups of a reference collection of specimens having spectral data of a spectral database and storing a trait of the unknown specimen in the spectral database, the method comprising:

• reducing a dimension of input spectral data of the spectral database of the reference collection using a data processor, the dimension reduction comprising principal component analysis;
• storing a first input trait of the unknown specimen in memory;
• receiving an input selecting a similarity metric of a plurality of different similarity metrics to classify the unknown specimen to the reference group of the reference collection of specimens;
• determining at least one similarity threshold value associated with the spectral data of the reference group of the reference collection via the data processor responsive to the selected similarity metric for classifying the unknown specimen to the reference group of the reference collection of specimens;
• and
• predicting a value of a second different trait of the unknown specimen by determining membership of the unknown specimen in the reference group of the reference collection of specimens, the spectral database storing the second different trait of the reference group of specimens for output once membership of the unknown specimen in the reference group is determined by the data processor.

A computer implemented method for classifying an unknown specimen having spectral data to a reference group of a plurality of different reference groups of a reference collection of specimens of a reference spectral database of a data processor, a reference specimen having a first trait, the method comprising:

• storing measured input spectral data for a reference group of a plurality of different groups of reference specimens in said reference spectral database of data processor memory, each reference spectral data specimen of the spectral database comprising spectral data of magnitude at a frequency and a value for the first trait associated with the reference group of the spectral database;
• reducing a dimension of the input spectral data, the dimension reduction comprising binning by spectral frequency;
• determining a similarity threshold for membership in the reference group of reference specimens using a selected metric of a plurality of different similarity metrics,
• classifying the unknown specimen having a different trait as belonging to the reference group of the reference spectral database using the selected similarity metric of the plurality of different similarity metrics;
• and associating the different trait of the unknown specimen as belonging to the group of reference specimens having the first trait.

A computer implemented method for classifying an unknown specimen having spectral data to a reference group of a plurality of different reference groups of a reference collection of specimens of a reference spectral database of a data processor, a reference specimen having a first trait, the method comprising:

• storing measured input spectral data for a reference group of a plurality of different groups of reference specimens in said reference spectral database of data processor memory, each reference spectral data specimen of the spectral database comprising spectral data of magnitude at a frequency and a value for the first trait associated with the reference group of the spectral database;
• reducing a dimension of the input spectral data, the dimension reduction comprising utilizing a projection selected to reveal structure inherent in the spectral data;
• determining a similarity threshold for membership in the reference group of reference specimens using a selected metric of a plurality of different similarity metrics,
• classifying the unknown specimen baying a different trait as belonging to the reference group of the reference spectral database using the selected similarity metric of the plurality of different similarity metrics;
• and associating the different trait of the unknown specimen as belonging to the group of reference specimens having the first trait.