Neural Network Based Pattern Recognition Research Articles

Neural Network-based Pattern Recognition in the Framework of Edge Computing

Neural network (NN) model has been widely used in pattern recognition (PR), speech recognition, image processing and other fields, but its application in edge computing (EC) environment faces performance and energy consumption problems. This article first introduced the basic structure and training process of NN, including backpropagation algorithms. Then, this article presented a NN modeling approach based on EC, including NN model compression, distributed NN model and knowledge distillation approach. Finally, this article implemented a PR model for the MNIST (Mixed National Institute of Standards and Technology database) dataset and analyzed the experimental results. The experimental outcomes indicated that the presented approach can significantly enhance the performance of the NN model in the EC environment, while ensuring a high recognition accuracy. The NN modeling approach based on EC can reduce the amount of computation and storage of the NN, thus improving the operating efficiency of the NN in the EC environment by 6%-12%. The NN modeling approach based on EC can optimize the performance and efficiency of the NN model in the EC environment, and provide new ideas and approaches for the application of NN in the EC environment.

Convolutional neural network-based pattern recognition in natural circulation instability images

Heat removal systems employing natural circulation are key in new nuclear power plant designs for mitigating accidents. This study applies Convolutional Neural Networks (CNNs) to classify 'chugging' instability phases, analyzing 1152 two-phase flow images from a Natural Circulation Circuit. Three CNN models, including one incorporating transfer learning from the ImageNet database, were trained via five-fold cross-validation to fine-tune hyperparameters. This involved comparing models with and without transfer learning against a baseline linear model. A model using a pre-trained Resnet50 with transfer learning accurately classified all 230 samples, outperforming the baseline linear model with an F1-Score of 0.859. The results endorse the use of CNNs with transfer learning for thermohydraulic image analysis in identifying natural circulation instability stages.

Improved Resistive and Synaptic Characteristics in Neuromorphic Systems Achieved Using the Double-Forming Process.

In this study, we investigate the electrical properties of ITO/ZrOx/TaN RRAM devices for neuromorphic computing applications. The thickness and material composition of the device are analyzed using transmission electron microscopy. Additionally, the existence of TaON interface layers was confirmed using dispersive X-ray spectroscopy and X-ray photoelectron analysis. The forming process of the ZrOx-based device can be divided into two categories, namely single- and double forming, based on the initial lattice oxygen vacancies. The resistive switching behaviors of the two forming methods are compared in terms of the uniformity properties of endurance and retention. The rationale behind each I-V forming process was determined as follows: in the double-forming method case, an energy band diagram was constructed using F-N tunneling; conversely, in the single-forming method case, the ratio of oxygen vacancies was extracted based on XPS analysis to identify the conditions for filament formation. Subsequently, synaptic simulations for the applications of neuromorphic systems were conducted using a pulse scheme to achieve potentiation and depression with a deep neural network-based pattern recognition system to display the achieved recognition accuracy. Finally, high-order synaptic plasticity (spike-timing-dependent plasticity (STDP)) is emulated based on the Hebbian rule.

Toward the Optimization of Mining Operations Using an Automatic Unmineable Inclusions Detection System for Bucket Wheel Excavator Collision Prevention: A Synthetic Study

This work introduces a methodology for the automatic unmineable inclusions detection and Bucket Wheel Excavator (BWE) collision prevention, using electromagnetic (EM) inspection and a fuzzy inference system. EM data are collected continuously ahead from the bucket wheel of a BWE and subjected to processing. Two distinct methodologies for data processing were developed and integrated into the MATLAB programming environment. The first approach, named “Simple Mode”, utilizes statistical process control to generate real-time alerts in the event of a potential collision involving the excavator’s bucket and hard rock inclusions. The advanced processing flow (“Advanced Mode”) requires accurate instrument positioning and data from successive EM scans. It incorporates techniques of local resistivity maxima detection (Position Prominence Index) as well as Neural Network-based Pattern Recognition (NNPR). A decision support process based on a Fuzzy Inference System (FIS) has been developed to assist BWE operators in avoiding collision when digging hard rock inclusions. The proposed methodology was extensively tested using synthetic EM data. Limited real data, acquired with a CMD2 (GF Instruments) EM instrument equipped with GPS, were used to control its efficiency. Increased accuracy in the automatic detection of unmineable inclusions was observed using the Advanced Mode. On the other hand, the Simple Mode processing technique offers the advantage of being independent of instrument positioning as well as it provides real-time inspection of the excavated mine slope. This work introduces a methodology for hard rock inclusion detection and can contribute to the optimization of mine operations by improving resource efficiency, safety, cost savings, and environmental sustainability.

Neural network based pattern recognition for classification of the forced and natural oscillation

Forced Oscillation (FO) has emerged as a major concern in power system due to its unpredictable frequency range. When the frequency of FO lies in the vicinity of that of the inherent natural oscillation, it can lead to resonance, posing a serious risk to the system. Improper identification and dissociation of FO in a power system can imprint huge detrimental effects, including blackouts. To address this issue, a novel method based on pattern recognition using artificial neural networks (ANNs) is proposed in this paper to classify FO and electromechanical oscillation. Multivariate time series data are collected from phasor measurement units (PMUs) placed at different buses are used as input for a two-layer NN, which is optimized using the stochastic scaled conjugate gradient algorithm. The proposed algorithm is simulated on the standard IEEE 10 machine, 39 bus system for various cases. The accuracy of the proposed algorithm is evaluated using the F1 score, obtained from the confusion plot. The results demonstrate that the network achieves a training accuracy of 97.5% and a testing accuracy of 96.67% with an F1 score of 96.7%. Additionally, ROC curves shows the efficiency of the designed model to predict the FO.

Fast convolutional neural networks for identifying long-lived particles in a high-granularity calorimeter

We present a first proof of concept to directly use neural network based pattern recognition to trigger on distinct calorimeter signatures from displaced particles, such as those that arise from the decays of exotic long-lived particles. The study is performed for a high granularity forward calorimeter similar to the planned high granularity calorimeter for the high luminosity upgrade of the CMS detector at the CERN Large Hadron Collider. Without assuming a particular model that predicts long-lived particles, we show that a simple convolutional neural network, that could in principle be deployed on dedicated fast hardware, can efficiently identify showers from displaced particles down to low energies while providing a low trigger rate.

Warships Formation Identification Based on Neural Network

Submarine tactical software should be able to identify warships formation. Neural network based pattern recognition method is introduced in submarine tactical software to improve accuracy. This paper discussed how to transform sensor data into training samples by using Matlab, and how to choose training samples. Experiments indicate that GRNN network is capable of this work and suitable for actualizing.

Microstructural engineering in interface-type synapse device for enhancing linear and symmetric conductance changes

The origins of the nonlinear and asymmetric synaptic characteristics of TiOx-based synapse devices were investigated. Based on the origins, a microstructural electrode was utilized to improve the synaptic characteristics. Under an identical pulse bias, a TiOx-based synapse device exhibited saturated conductance changes, which led to nonlinear and asymmetric synaptic characteristics. The formation of an interfacial layer between the electrode and TiOx layer, which can limit consecutive oxygen migration and chemical reactions, was considered as the main origin of the conductance saturation behavior. To achieve consecutive oxygen migration and chemical reactions, structural engineering was utilized. The resultant microstructural electrode noticeably improved the synaptic characteristics, including the unsaturated, linear, and symmetric conductance changes. These synaptic characteristics resulted in the recognition accuracy significantly increasing from 38% to 90% in a neural network-based pattern recognition simulation.

Characterization of cytoplasmic viscosity of hundreds of single tumour cells based on micropipette aspiration.

Cytoplasmic viscosity (μc) is a key biomechanical parameter for evaluating the status of cellular cytoskeletons. Previous studies focused on white blood cells, but the data of cytoplasmic viscosity for tumour cells were missing. Tumour cells (H1299, A549 and drug-treated H1299 with compromised cytoskeletons) were aspirated continuously through a micropipette at a pressure of −10 or −5 kPa where aspiration lengths as a function of time were obtained and translated to cytoplasmic viscosity based on a theoretical Newtonian fluid model. Quartile coefficients of dispersion were quantified to evaluate the distributions of cytoplasmic viscosity within the same cell type while neural network-based pattern recognitions were used to classify different cell types based on cytoplasmic viscosity. The single-cell cytoplasmic viscosity with three quartiles and the quartile coefficient of dispersion were quantified as 16.7 Pa s, 42.1 Pa s, 110.3 Pa s and 74% for H1299 cells at −10 kPa (ncell = 652); 144.8 Pa s, 489.8 Pa s, 1390.7 Pa s, and 81% for A549 cells at −10 kPa (ncell = 785); 7.1 Pa s, 13.7 Pa s, 31.5 Pa s, and 63% for CD-treated H1299 cells at −10 kPa (ncell = 651); and 16.9 Pa s, 48.2 Pa s, 150.2 Pa s, and 80% for H1299 cells at −5 kPa (ncell = 600), respectively. Neural network-based pattern recognition produced successful classification rates of 76.7% for H1299 versus A549, 67.0% for H1299 versus drug-treated H1299 and 50.3% for H1299 at −5 and −10 kPa. Variations of cytoplasmic viscosity were observed within the same cell type and among different cell types, suggesting the potential role of cytoplasmic viscosity in cell status evaluation and cell type classification.

Mechanical property characterization of hundreds of single nuclei based on microfluidic constriction channel.

As label-free biomarkers, the mechanical properties of nuclei are widely treated as promising biomechanical markers for cell type classification and cellular status evaluation. However, previously reported mechanical parameters were derived from only around 10 nuclei, lacking statistical significances due to low sample numbers. To address this issue, nuclei were first isolated from SW620 and A549 cells, respectively, using a chemical treatment method. This was followed by aspirating them through two types of microfluidic constriction channels for mechanical property characterization. In this study, hundreds of nuclei were characterized, producing passage times of 0.5 ± 1.2 s for SW620 nuclei in type I constriction channel (n = 153), 0.045 ± 0.047 s for SW620 nuclei in type II constriction channel (n = 215) and 0.50 ± 0.86 s for A549 nuclei in type II constriction channel. In addition, neural network based pattern recognition was used to classify the nuclei isolated from SW620 and A549 cells, producing successful classification rates of 87.2% for diameters of nuclei, 85.5% for passage times of nuclei and 89.3% for both passage times and diameters of nuclei. These results indicate that the characterization of the mechanical properties of nuclei may contribute to the classification of different tumor cells.

Membrane capacitance of thousands of single white blood cells.

As label-free biomarkers, the electrical properties of single cells are widely used for cell type classification and cellular status evaluation. However, as intrinsic cellular electrical markers, previously reported membrane capacitances (e.g. specific membrane capacitance Cspec and total membrane capacitance Cmem) of white blood cells were derived from tens of single cells, lacking statistical significance due to low cell numbers. In this study, white blood cells were first separated into granulocytes and lymphocytes by density gradient centrifugation and were then aspirated through a microfluidic constriction channel to characterize both Cspec and Cmem Thousands of granulocytes (ncell = 3327) and lymphocytes (ncell = 3302) from 10 healthy blood donors were characterized, resulting in Cspec values of 1.95 ± 0.22 µF cm-2 versus 2.39 ± 0.39 µF cm-2 and Cmem values of 6.81 ± 1.09 pF versus 4.63 ± 0.57 pF. Statistically significant differences between granulocytes and lymphocytes were located for both Cspec and Cmem In addition, neural network-based pattern recognition was used to classify white blood cells, producing successful classification rates of 78.1% for Cspec and 91.3% for Cmem, respectively. These results indicate that as intrinsic bioelectrical markers, membrane capacitances may contribute to the classification of white blood cells.

Energy-efficient Systems for the Sensing and Separation of Mixed Polymers

Polymers are ubiquitous in modern manufactured products. The potential detrimental impacts of their end-of-life disposal have stimulated significant increases in recycling rates. Recyclate purity is paramount; however this must be achieved with a positive net energy balance. Existing technologies for identification and separation of polymers are often both expensive and energy intensive. This paper investigates Infrared (IR) imaging to extract information on thermal properties of various product polymers within a recycling line. An intelligent decision making support system is enabled using neural network based pattern recognition for automatic polymer identification and classification. Potential energy savings versus current technologies are discussed.

A new security approach for public transport application against tag cloning with neural network-based pattern recognition

RFID tags are widely used in situations where their counterfeiting or cloning can bring financial rewards. Cloning is a particular problem because it gets round the sophisticated security measures. This paper describes a neural network-based technique for identifying cloned tickets for a public transport system. It is based on modeling passenger behavior. Cardholders' behavioral characteristics in using public transport are modeled with seven neural network model equations, one for each day of the week, and stored in an RFID card. At the time of use, these model equations or characteristics are employed to predict whether the user is the real owner of the card. Therefore, even if the RFID card is cloned, the cloned card cannot be used because a passenger's behavioral characteristics when using public transport are individual and unique, such as the passenger's signature or style of speech. Therefore, the proposed approach provides high security, especially for low-cost RFID tags.

Classification of Radionuclides on Polyvinyl Toluene Radiation Portal Monitors by a Neural Network Based System§

To detect radiation sources in containers, radiation portal monitors (RPMs) are normally used. Among the various types of RPM, polyvinyl toluene (PVT) monitors are the most widely used, thanks to their lower cost. However, they provide only low resolution when energy spectra are measured. For this reason, it is currently not possible to identify what kind of radiation source is being measured, only that there is some radiation source present, when an alarm is triggered in a PVT monitor. In this case, a second inspection is necessary in order to distinguish between innocent alarms and those due to potentially dangerous radioactive material. This second inspection involves a considerable outlay of money and time at frontiers. The main purpose of this paper is to propose an algorithm which can be applied so as to reduce the number of innocent alarms purely on the basis of the data obtained by the PVT monitors. To this end, an intelligent system based on two stages is proposed: a preprocessing of the energy spectrum and a neural network based pattern recognition module, which classifies the radiation source into one of the standard groups. This intelligent system has been tested with real radiation sources, whose energy spectra were measured with two different PVT detectors, yielding good results.

Residual Stress Condition Monitoring via Sensor Fusion in Turning of Inconel 718

A multiple sensor monitoring system, equipped with cutting force, acoustic emission and vibration sensing units, was employed in association with advanced procedures for signal analysis, sensor fusion and cognitive decision making for residual stress evaluation in turning of Inconel 718 nickel alloy. Two signal processing and feature extraction methodologies, based respectively on sensory data statistical evaluation and Principal Component Analysis, were applied to the sensor signals generated during experimental turning tests. The extracted features were combined into sensor fusion input feature vectors to be fed to neural network based pattern recognition paradigms for decision making on machined surface integrity in terms of residual stress conditions.

A microfluidic system for cell type classification based on cellular size-independent electrical properties

This paper presents a microfluidic system enabling cell type classification based on continuous characterization of size-independent electrical properties (e.g., specific membrane capacitance (C(specific membrane)) and cytoplasm conductivity (σ(cytoplasm)). In this study, cells were aspirated continuously through a constriction channel, while cell elongation and impedance profiles at two frequencies (1 kHz and 100 kHz) were measured simultaneously. Based on a proposed distributed equivalent circuit model, 1 kHz impedance data were used to evaluate cellular sealing properties with constriction channel walls and 100 kHz impedance data were translated to C(specific membrane) and σ(cytoplasm). Two lung cancer cell lines of CRL-5803 cells (n(cell) = 489) and CCL-185 cells (n(cell) = 487) were used to evaluate this technique, producing a C(specific membrane) of 1.63 ± 0.52 μF cm(-2) vs. 2.00 ± 0.60 μF cm(-2), and σ(cytoplasm) of 0.90 ± 0.19 S m(-1)vs. 0.73 ± 0.17 S m(-1). Neural network-based pattern recognition was used to classify CRL-5803 and CCL-185 cells, producing success rates of 65.4% (C(specific membrane)), 71.4% (σ(cytoplasm)), and 74.4% (C(specific membrane) and σ(cytoplasm)), suggesting that these two tumor cell lines can be classified based on their electrical properties.

Automatic segmentation and supervised learning-based selection of nuclei in cancer tissue images.

Analysis of preferential localization of certain genes within the cell nuclei is emerging as a new technique for the diagnosis of breast cancer. Quantitation requires accurate segmentation of 100-200 cell nuclei in each tissue section to draw a statistically significant result. Thus, for large-scale analysis, manual processing is too time consuming and subjective. Fortuitously, acquired images generally contain many more nuclei than are needed for analysis. Therefore, we developed an integrated workflow that selects, following automatic segmentation, a subpopulation of accurately delineated nuclei for positioning of fluorescence in situ hybridization-labeled genes of interest. Segmentation was performed by a multistage watershed-based algorithm and screening by an artificial neural network-based pattern recognition engine. The performance of the workflow was quantified in terms of the fraction of automatically selected nuclei that were visually confirmed as well segmented and by the boundary accuracy of the well-segmented nuclei relative to a 2D dynamic programming-based reference segmentation method. Application of the method was demonstrated for discriminating normal and cancerous breast tissue sections based on the differential positioning of the HES5 gene. Automatic results agreed with manual analysis in 11 out of 14 cancers, all four normal cases, and all five noncancerous breast disease cases, thus showing the accuracy and robustness of the proposed approach.

Impact of gender on snore-based obstructive sleep apnea screening

Obstructive sleep apnea syndrome (OSA) is a serious widespread disease in which upper airways (UA) are collapsed during sleep. OSA has marked male predominance in prevalence. Although women are less vulnerable to OSA, under-diagnosed OSA in women may associate with serious consequences. Snoring is commonly associated with OSA and one of the earliest symptoms. Snore sounds (SS) are generated due to vibration of the collapsing soft tissues of the UA. Structural and functional properties of the UA are gender dependent. SS capture these time varying gender attributed UA properties and those could be embedded in the acoustic properties of SS. In this paper, we investigate the gender-specific acoustic property differences of SS and try to exploit these differences to enhance the snore-based OSA detection performance. We developed a snore-based multi-feature vector for OSA screening and one time-measured neck circumference was augmented. Snore features were estimated from SS recorded in a sleep laboratory from 35 females and 51 males and multi-layer neural network-based pattern recognition algorithms were used for OSA/non-OSA classification. The results were K-fold cross-validated. Gender-dependent modeling resulted in an increase of around 7% in sensitivity and 6% in specificity at the decision threshold AHI = 15 against a gender-neutral model. These results established the importance of adopting gender-specific models for the snore-based OSA screening technique.

Sensor Fusion for Tool State Classification in Nickel Superalloy High Performance Cutting

A multiple sensor monitoring system, endowed with cutting force, acoustic emission and vibration sensing units, was employed for tool state classification in turning of Inconel 718. A sensor fusion signal processing paradigm based on the Principal Component Analysis was applied to the sensor signals generated during cutting in order to reduce the high dimensionality of the sensory data by extracting significant signal features. The principal components, obtained through Principal Component Analysis of sensor fusion data matrices and strongly related to sensor signals, were used as input features to a neural network based pattern recognition procedure for decision making on tool wear condition.

Investigating Neural Correlates of Behavior in Freely Behaving Rodents Using Inertial Sensors

Simultaneous behavior and multielectrode neural recordings in freely behaving rodents holds great promise to study the neural bases of behavior and disease models in combination with genetic manipulations. Here, we introduce the use of three-axis accelerometers to characterize the behavior of rats and mice during chronic neural recordings. These sensors were small and light enough to be worn by rodents and were used to record three-axis acceleration during freely moving behavior. A two-layer neural network-based pattern recognition algorithm was developed to extract the natural behavior of mice from the acceleration data. Successful recognition of resting, eating, grooming, and rearing are shown using this approach. The inertial sensors were combined with continuous 24-h recordings of neural data from the striatum of mice to characterize variations in neural activity with circadian cycles and to study the neural correlates of spontaneous action initiation. Finally, accelerometers were used to study the performance of rodents in traditional operant conditioning, where they were used to extract the reaction time of rodents. Thus the addition of accelerometer recordings of rodents to chronic multielectrode neural recordings provides great value for a number of neuroscience applications.