Chemical sensor pattern recognition system and method using a self-training neural network classifier with automated outlier detection

A method of pattern classification using a probabilistic neural network comprising the steps of:

• receiving sensory data from partially selective sensors to be used by a pattern extraction unit in generating pattern vectors;
• comparing to determine a selection between a set of initial pattern vectors contained in an initial hidden layer of an initial probabilistic neural network to a set of current neurons in a hidden layer of a probabilistic neural network obtained after processing said pattern vectors, said current set of neurons obtained by using a learning vector quantization procedure based on an Euclidean distance;
• identifying which neurons of said set of current neurons are the nearest to each pattern vector of said set of pattern vectors;
• and
• eliminating from said probabilistic neural network those neurons of said set of current neurons which have not been identified as being nearest to any of said set of pattern vectors so that the remaining set of current neurons are selected to be used for pattern classification and so that said probabilistic neural network provides a corresponding output thereof for determination of said pattern classification for identifying chemical analytes.

A method of pattern classification used to identify chemical analytes comprising the steps of;

• receiving a set of sensory data from a plurality of partially selective sensors;
• extracting a set of pattern vectors from said set of sensory data which form a set of initial hidden layer neurons of an initial probabilistic neural network;
• reducing the number of neurons in said set of initial hidden layer neurons through learning vector quantization to form a set of hidden layer neurons in said probabilistic neural network;
• identifying neurons in said set of hidden layer neurons, already having been reduced by said reducing step, which have the shortest Euclidean distance to said set of extracted pattern vectors;
• eliminating neurons in said set of hidden layer neurons of said identifying step which have not been identified as having the shortest Euclidean distance to said pattern vectors;
• determining an optimal global σ value for all neurons in said probabilistic neural network after said elimination of said neurons in said set of hidden layer neurons, wherein said optimal global σ value is computed from the summation of dot product distances between each neuron and its nearest neighbor;
• calculating an outlier rejection threshold value based on said optimal global σ value;
• generating a set of random pattern vectors using a Monte Carlo simulation;
• and
• adjusting said outlier threshold value so that a predetermined percentage of said random pattern vectors are rejected as outliers.

A neural network pattern classifier used for identifying chemical analytes using a probabilistic neural network comprising:

• a plurality of partially selective sensors to generate sensory data in response to the presence of chemical analytes;
• a pattern extraction unit to generate a plurality of pattern vectors using said sensory data;
• a probabilistic neural network receiving said plurality of pattern vectors and forming them into an initial hidden layer of pattern vectors;
• a dead neuron elimination unit receiving said initial hidden layer of pattern vectors and processing the initial hidden layer of pattern vectors to determine a current set of neurons and identifying which neurons in the current set of neurons are nearest to each pattern vector of said pattern vectors of said initial hidden layer, and eliminating those neurons of said current set of neurons that have not been identified as being nearest to any pattern vector of said pattern vectors of said initial hidden layer, so that the remaining set of neurons are selected to be used for pattern classification so that said dead neuron elimination unit provides a corresponding output thereof for determination of said pattern classification for identifying chemical analytes.

A probabilistic neural network training computer system used to identify chemical analytes;

• comprising:
• a plurality of partially selective sensors to generate a set of sensory data when exposed to said chemical analytes;
• a pattern extraction unit connected to said plurality of partially selective sensors to extract a set of pattern vectors from said set of sensory data which form a set of hidden layer neurons of a probabilistic neural network;
• a learning vector quantization classifier unit connected to said pattern extraction unit to reduce the number of neurons in said set of hidden layer neurons using learning vector quantization and to provide a corresponding output thereof;
• a dead neuron elimination unit connected to receive said output of said learning vector quantization classifier unit to identify neurons in said set of hidden layer neurons which have the shortest Euclidean distance to said set of pattern vectors and eliminate neurons in said set of hidden layer neurons which have not been identified as having the shortest Euclidean distance to any pattern vectors of said set of pattern vectors, said dead neuron elimination unit providing an output representative of said set of hidden layer neurons having the shortest Euclidean distance;
• a σ optimization unit connected to receive said output of said dead neuron elimination unit and to determine an optimal global σ value for all neurons in said probabilistic neural network, wherein said optimal global σ value is computed from the summation of dot product distances between each neuron and its nearest neighbor and said σ optimization unit provides a corresponding output;
• a threshold determination unit connected to receive said output of said σ optimization unit and to calculate an outlier rejection threshold value based on said optimal global σ value and said threshold determination unit provides a corresponding output;
• a Monte Carlo simulation unit connected to receive said output of said threshold determination unit and to generate a set of random pattern vectors using a Monte Carlo simulation;
• and
• said threshold determination unit in cooperation with said Monte Carlo simulation unit adjusting said outlier threshold value so that a predetermined percentage of said random pattern vectors rejected as outliers.