Neural Network Pattern Recognition System Research Articles

Rule extraction from the Artificial Neural Network

Applying machine learning techniques to various critical decision-making domains implies the need to explain why a machine learning algorithm makes certain conclusions. Artificial neural networks (ANNs) have achieved high classification accuracy for many classification problems, but their results can’t be interpreted, which is why ANNs are often considered as «black box». To interpret the results of neural network classification, the paper presents a method for extracting rules from ANN. The proposed method is based on the structuring of information flows processed in the ANN information field through transforming complex multidimensional data into a simpler structure of a lower dimension, followed by a trivial transformation of the results into a set of fuzzy rules of a certain type. The implementation of the proposed method provides for the construction of the hybrid ANN configuration based on self-organizing and multilayer ANNs. The results of an experimental research of the proposed method are presented on the example of solving the well-known problem of multiparameter classification. The results obtained confirm the adequacy of the proposed method, which can be used both in independent neural network pattern recognition systems and within decision support systems.

The detection of organic solvent vapor by using polymer coated chemocapacitor sensor

A chemocapacitor consists of planar interdigital electrodes (IDE) made by two comb electrodes on a substrate. A dielectric film was applied on the electrodes in which the absorbed vapor will modify its permittivity. This study has fabricated chemocapacitor with the IDE distance of 0.5 mm, while the dielectric film was a sensitive layer consisting of a polymeric material. The deposition of the polymeric film was accomplished by drop casting. A sensor array consisting of four chemocapacitors coated with different polymers namely PEG-1540, PEG-20M, PEG-6000, and PVP was used to obtain the pattern of shift in the capacitance. The integrated circuit AD7746 was used as the capacitance to-digital converter (CDC). The organic solvents of ethanol, benzene, and aceton were used as the vapor samples in this experiment. The results showed that the change in the capacitance value increases proportionally to the concentration of vapour where sensors coated with PEG-1540 and PVP have higher sensitivity, i.e. 0.0028pF/part per thousand and 0.0027pF/part per thousand, respectively. Based on the capacitance to digital conversion capabilities, the system provides there solution of 0.4084ppm. The sensor array could produce a different pattern for each of the vapor sample. The Neural Network pattern recognition system could identify the type of vapor automatically with the root mean square error of 10-5

Development of the decision support subsystem in the systems of neural network pattern recognition by statistical information

The decision support subsystem in the neural network pattern recognition system, allowing to reduce the subjectivity and increase the quality of expert decisions in the construction of training samples by the statistical data of observations over the state of the management facilities of the production or social environment is developed. A functional structure of the NNPR system of the management facilities of the production or social environment by statistical information is proposed. According to the proposed structure, the problem-oriented NNPR system consists of subsystems of the initial data preparation for the PR and storage of the aggregated data, neural network models, and knowledge about the management facility of the production or social environment, as well as the support of the decision-making about its state. The decision support subsystem based on the proposed method of support of the classification decision-making using the Kohonen self-organizing layer is developed. The developed method of support of the classification decision-making provides a consistent implementation of the stages of self-organizing of neurons of the Kohonen computing layer, the calibration of the elements of the output vector of the training sample and their final labeling. An example of implementation of the proposed subsystem of support of decision-making about the class of the initialized labor protection project based on the expert estimates of the current level of the organization and working conditions at the enterprise is given. Using the developed decision support subsystem in the systems of NNPR by statistical information allows to adjust the decisions of the DM in accordance with the decisions of the neural network output machine, and thereby increase a relative share of correct expert estimates by 20 % on average and reduce false estimates for a number of pattern recognition problems by 50 %.

Photopolarimetric lidar duel-beam switching device and Mueller matrix standoff detection method

We present an optomechanical switching device (OSD) for a photopolarimetric lidar system with differential-absorption Mueller matrix spectroscopy standoff detection method. An output train of alternate continuous-wave CO2 laser beams [...L1:L2...] is directed onto suspect chemical-biological (CB) aerosol plume or the land mass it contaminates (S) vis-à-vis the OSD, where L1 [L2] is tuned on [detuned off] a resonant molecular absorption moiety of CB analyte. Moreover, both incident beams and their backscattered radiances, from S, are polarization-modulated synchronously so as to produce gated temporal voltage waveforms called scattergrams recorded on focus at the receiver end of polarization lidar sensor system. All 16 elements of the Mueller matrix (Mij) of S are measured via digital or analog filtration of constituent frequency components in these running scattergram data streams (phase-sensitive detection). A collective set of normalized differential elements {Mi,j} (ratioed to element M11) that are susceptible to the analyte, probed on-then-off its molecular absorption band, form a unique detection domain that is scrutinized. Any mapping onto this domain in Mueller-space, from incoming sensor scattergram data sets preprocessed by algorithm and forwarded through a trained neural network pattern recognition system, cues a standoff detection event.

Neural network pattern recognition analysis of graft flow characteristics improves intra-operative anastomotic error detection in minimally invasive CABG.

The intra-operative assessment of the quality of anastomosis in minimally invasive coronary artery bypass surgery (CABG) is critical. Recent investigations demonstrated that flow probes used intra-operatively to assess anastomotic errors may give the surgeon a false sense of confidence as only severely stenotic anastomoses (>90%) could be reliably detected. We developed a neural network system using graft flow data and assessed its potential to improve anastomotic error detection. Mammary to LAD grafts (n = 46) were constructed in mongrel dogs off-pump. Continuous beat-to-beat graft flow was recorded using transit-time flow probes. Various degrees of anastomotic stenoses (0-100%) were created by an additional suture. The degree of anastomotic stenosis was confirmed by postoperative angiography. A learning vector quantization neural network was created using heart rate, mean aortic pressure, mean systolic, maximum systolic, minimum systolic, mean diastolic, maximum diastolic, minimum diastolic, and mean graft flows. In addition, a spectral analysis of the flow waveforms was performed and the magnitude and phase of the first five harmonics were used to further develop the neural network. The neural network pattern recognition system was 94% accurate in detecting any stenosis >50%. To validate the model, a testing set was used with 20% of the data values, and the accuracy remained at 100% above chance alone. Pattern recognition of transit-time flow probe tracings using neural network systems can detect anastomotic errors significantly better than the surgeon's visual assessment, thereby improving the clinical outcome of minimally invasive CABG.

Differential absorption Mueller matrix spectroscopy and the infrared detection of crystalline organics

The complete 16-element Mueller matrices for backscattering from amino acids, sugars, and other enantiomorphic compounds pressed into wafer form were measured at infrared wavelengths. For each compound a pair of CO(2) laser lines was selected from the 9.1-11.6-mum region such that one line excited an absorption band in the compound, whereas the other did not. It was observed that at least some of the matrix elements differed significantly depending on which of the two wavelengths was used in the measurement. We propose that a neural network pattern recognition system can be trained to detect the presence of specific compounds based on multiwavelength backscatter Mueller matrix measurements.

Knowing That You Know: An Essay on Mechanistic Metaknowledge

Humans are adept at providing accurate statements of confidence in their perceptual identification and recall memory responses. In spite of this, mechanical pattern-recognition systems and other artificial intelligence devices seldom express response certainty. The purpose of this paper is to show how useful confidence ratings can be in integrating the results of a variety of pattern-recognition systems to produce a single, optimal decision concerning the target to be recognized. We outline several ways neural network pattern-recognition systems could be modified to issue confidence ratings with each classification response. In sketching a mechanical system of confidence ratings we find we have also provided a preliminary framework for understanding human confidence judgments and human metacognition.

KNOWING THAT YOU KNOW: AN ESSAY ON MECHANISTIC METAKNOWLEDGE

Humans are adept at providing accurate statements of confidence in their perceptual identification and recall memory responses. In spite of this, mechanical pattern-recognition systems and other artificial intelligence devices seldom express response certainty. The purpose of this paper is to show how useful confidence ratings can be in integrating the results of a variety of pattern-recognition systems to produce a single, optimal decision concerning the target to be recognized. We outline several ways neural network pattern-recognition systems could be modified to issue confidence ratings with each classification response. In sketching a mechanical system of confidence ratings we find we have also provided a preliminary framework for understanding human confidence judgments and human metacognition.