METHOD OF PROCESSING VECTOR SIGNALS FOR PATTERN RECOGNITION BASED ON NEURAL NETWORKS WITH DETECTION OF OUTBURSTS Russian patent published in 2024 - IPC G06N20/10 G01N35/00 G06T1/40 

FIELD: data processing.

SUBSTANCE: method of processing vector signals for pattern recognition based on neural networks with detecting outbursts includes: A) obtaining an array of samples of the measured signal to form a training sample, B) primary processing of the measured one-dimensional signal x(t) by means of: i) removing low frequency noise; ii) local correction of the baseline within the regions of the stationary regions of low temperature. At step C) performing the subsequent normalization of the preprocessed signal; D) normalized according to cl. C) the sample is then projected onto the account space S of the model by PLS; E) for each sample, its relevance to the problem being solved is evaluated, F) frequent observation of "outbursts" according to cl. E) indicates systematic problems of the measuring apparatus or mathematical model; G) if the sample satisfies the criterion of cl. E), then it is subjected to so-called "min-max" normalization, H) sample subjected to treatment according to cl. G), is supplied to the input of the neural network; I) the value obtained as a result of using the neural network according to cl. H) is subjected to inverse "min-max" normalization, by multiplying by the maximum concentration value ymax(t) and adding the minimum value ymin(t) of hydrogen concentrations (scalar values ymin(t) and ymax(t) of hydrogen concentrations are calculated in advance based on the training sample of examples according to cl. A), which is the desired target hydrogen concentration.

EFFECT: high accuracy of signal processing.

15 cl, 2 dwg