Neural Network Pattern Recognition Research Articles

Application of Neural Networks for Pattern Recognition

This article details the process of creating and training a neural network for the task of classifying dog and cat images using TensorFlow and the MobileNetV2 architecture. Data preparation and preprocessing, including image resizing and normalization, are described. Details of the integration of the pre-trained MobileNetV2 model are given, the process of pre-training the model on specific data is demonstrated, as well as methods of model construction and optimization, including the addition and tuning of additional layers. Special attention is paid to practical application of the trained model for classification of new images, including loading, processing, prediction and visualization of results.

Investigation of the acoustic properties of underwater rainfall noise measured by a bottom-mounted hydrophone and its application to deep learning to classify and estimate rainfall

Although the ocean covers more than 70 % of the Earth's surface, it is difficult to measure precipitation over the ocean due to the technical limitations of platforms such as surface buoys and satellites. Recently, the estimation of precipitation has been tried using the underwater sound generated by rainfall. In this talk, we present the results of comparing underwater noise and rainfall with and without precipitation using the underwater noise and rainfall data provided by OOI (Ocean Observatories Initiative, USA). The acoustic data was acquired by a bottom-moored hydrophone at a depth of 80 m off the coast of Oregon, and the rainfall data was measured from a rain gauge installed on a surface buoy. The 1-minute averaged power spectral densities for ambient noise were computed for 26 days from 5 - 30 April 2018, which were compared with rain rate and wind speed. Based on the comparison results, the presence or absence of rainfall was determined through a pattern recognition neural network, and finally, the amount of precipitation was estimated through multiple linear regression analysis.

Artificial visual‐tactile perception array for enhanced memory and neuromorphic computations

AbstractThe emulation of human multisensory functions to construct artificial perception systems is an intriguing challenge for developing humanoid robotics and cross‐modal human–machine interfaces. Inspired by human multisensory signal generation and neuroplasticity‐based signal processing, here, an artificial perceptual neuro array with visual‐tactile sensing, processing, learning, and memory is demonstrated. The neuromorphic bimodal perception array compactly combines an artificial photoelectric synapse network and an integrated mechanoluminescent layer, endowing individual and synergistic plastic modulation of optical and mechanical information, including short‐term memory, long‐term memory, paired pulse facilitation, and “learning‐experience” behavior. Sequential or superimposed visual and tactile stimuli inputs can efficiently simulate the associative learning process of “Pavlov's dog”. The fusion of visual and tactile modulation enables enhanced memory of the stimulation image during the learning process. A machine‐learning algorithm is coupled with an artificial neural network for pattern recognition, achieving a recognition accuracy of 70% for bimodal training, which is higher than that obtained by unimodal training. In addition, the artificial perceptual neuron has a low energy consumption of ∼20 pJ. With its mechanical compliance and simple architecture, the neuromorphic bimodal perception array has promising applications in large‐scale cross‐modal interactions and high‐throughput intelligent perceptions.image

Monitoring of polycyclic aromatic hydrocarbons and toxic metals in blue mussels and oysters for human consumption from Danish waters (2005–2019)

Bivalve molluscs have for many years been used as indicator organisms for environmental pollution with various chemical substances. However, limited information is available on the levels of chemical contaminants in mussels and oysters intended for human consumption. Polycyclic aromatic hydrocarbons (PAHs) and toxic metals, such as cadmium (Cd), mercury (Hg) and lead (Pb) have been analysed in edible bivalve samples and the present study provides the results from samples collected in Danish waters from 2005 to 2019. Principal Component Analysis (PCA) and Pattern Recognition Neural Network (PRNN) analysis were applied to data to identify any differences according to organism, locality, and contaminant levels. Differences were confirmed for Pb concentrations being generally higher in mussels than oysters. Based on toxic metals results, the PCA and PRNN analyses made it possible to categorise samples as a mussel or an oyster, as well as make decisions on the origin of a specific sample. All concentrations in mussels and oysters were below the current European Union maximum levels for the contaminants.

The Enhanced Performance of Neuromorphic Computing Hardware in an ITO/ZnO/HfOx/W Bilayer-Structured Memory Device.

This study discusses the potential application of ITO/ZnO/HfOx/W bilayer-structured memory devices in neuromorphic systems. These devices exhibit uniform resistive switching characteristics and demonstrate favorable endurance (>102) and stable retention (>104 s). Notably, the formation and rupture of filaments at the interface of ZnO and HfOx contribute to a higher ON/OFF ratio and improve cycle uniformity compared to RRAM devices without the HfOx layer. Additionally, the linearity of potentiation and depression responses validates their applicability in neural network pattern recognition, and spike-timing-dependent plasticity (STDP) behavior is observed. These findings collectively suggest that the ITO/ZnO/HfOx/W structure holds the potential to be a viable memory component for integration into neuromorphic systems.

Explainable AI approach with original vegetation data classifies spatio-temporal nitrogen in flows from ungauged catchments to the Great Barrier Reef

Transfer of processed data and parameters to ungauged catchments from the most similar gauged counterpart is a common technique in water quality modelling. But catchment similarities for Dissolved Inorganic Nitrogen (DIN) are ill posed, which affects the predictive capability of models reliant on such methods for simulating DIN. Spatial data proxies to classify catchments for most similar DIN responses are a demonstrated solution, yet their applicability to ungauged catchments is unexplored. We adopted a neural network pattern recognition model (ANN-PR) and explainable artificial intelligence approach (SHAP-XAI) to match all ungauged catchments that flow to the Great Barrier Reef to gauged ones based on proxy spatial data. Catchment match suitability was verified using a neural network water quality (ANN-WQ) simulator trained on gauged catchment datasets, tested by simulating DIN for matched catchments in unsupervised learning scenarios. We show that discriminating training data to DIN regime benefits ANN-WQ simulation performance in unsupervised scenarios ( p< 0.05). This phenomenon demonstrates that proxy spatial data is a useful tool to classify catchments with similar DIN regimes. Catchments lacking similarity with gauged ones are identified as priority monitoring areas to gain observed data for all DIN regimes in catchments that flow to the Great Barrier Reef, Australia.

The KI-ASIC Dataset

We present a novel dataset captured from a BMW X5 test carrier within the German research project KI-ASIC for use in radar sensor development and autonomous driving research. Our work aims at providing a blueprint for the process of creating labeled datasets for the development of neural networks for pattern recognition in radar data in the automotive environment. With a variety of different sensor types such as wide angle color cameras, a high-resolution color stereo camera, an Ouster OS1-64 laser scanner and three novel Infineon radar sensors, we recorded over 100,000 scenes of real traffic scenarios as well as defined test scenarios with a frequency of 10 Hz. The scenarios in real traffic contain inner-city situations, but also scenes from rural areas with static and dynamic objects. Besides, the defined test scenarios are based on the NCAP scenarios and focus mostly on turning, overtaking and follow-up maneuvers. The data from the different sensors is calibrated, synchronized and timestamped including raw and rectified information. Our dataset also contains labels for all detected objects from a defined class list with distance and angle properties. The content of the paper aims at the description of the recording test carrier, the format of the provided sensor data and the structure of the overall dataset

Supervised learning and pattern recognition in photonic spiking neural networks based on MRR and phase-change materials

Traditional computers are limited by the separation of memory and processor units, is difficult to achieve fast, efficient, and low-power computing. While photonic spiking neural networks (SNNs) can overcome these shortcomings, they encounter limitations in large-scale integration. Silicon photonics platform, compatible with mature Complementary Metal Oxide Semiconductor (CMOS) platforms, is a promising candidate for realizing large-scale photonic SNNs. In this work, we proposed an integrated photonic SNN by exploiting the photonic properties of phase-change material (PCM) Ge2Sb2Te5 (GST) and micro-ring resonators (MRR), and demonstrated its integrate-and-fire (IF) behavior. Based on a system-level behavioral model, we adopt an improved Tempotron-like ReSuMe supervised learning algorithm to train the proposed photonic SNNs and complete a pattern recognition task for the clock's 12 clockwise directions. Then the influence of different noise levels is considered, and the accuracy is close to 1 when the noise level is less than 0.2. We propose a photonic implementation of such an SNN system, and use wavelength division multiplexing to achieve a scalable architecture for the pattern recognition task. The collaborative design and optimization of hardware architecture and algorithm are realized, providing a theoretical basis for the realization of photonic SNN based on MRRs and PCM.

A Comprehensive Pattern Recognition Neural Network for Collision Classification Using Force Sensor Signals

In this paper, force sensor signals are classified using a pattern recognition neural network (PRNN). The signals are classified to show if there is a collision or not. In our previous work, the joints positions of a 2-DOF robot were used to estimate the external force sensor signal, which was attached at the robot end-effector, and the external joint torques of this robot based on a multilayer feedforward NN (MLFFNN). In the current work, the estimated force sensor signal and the external joints’ torques from the previous work are used as the inputs to the proposed designed PRNN, and its output is whether a collision is found or not. The designed PRNN is trained using a scaled conjugate gradient backpropagation algorithm and tested and validated using different data from the training one. The results prove that the PRNN is effective in classifying the force signals. Its effectiveness for classifying the collision cases is 92.8%, and for the non-collisions cases is 99.4%. Therefore, the overall efficiency is 99.2%. The same methodology and work are repeated using a PRNN trained using another algorithm, which is the Levenberg–Marquardt (PRNN-LM). The results using this structure prove that the PRNN-LM is also effective in classifying the force signals, and its overall effectiveness is 99.3%, which is slightly higher than the first PRNN. Finally, a comparison of the effectiveness of the proposed PRNN and PRNN-LM with other previous different classifiers is included. This comparison shows the effectiveness of the proposed PRNN and PRNN-LM.

An intelligent pattern recognition model for assessment of terrorists’ activities in Nigeria

Terrorism and its brutal tendencies constitute a major setback to the development process of the Nigerian economy leading to severe loss of lives, destruction of properties, and a decline of interest in investment by both local and foreign investors. Many models for assessment of terrorist’s activities lack the ability of learning from previous patterns in order to guide pre-emptive actions against future occurrences, and there are no established regional pattern of weaponry, types of attack, as well as types of victims of terrorists’ operations. This study seeks to build a robust intelligent model for recognizing several terrorists’ patterns in each of the six geo-political zones of Nigeria. A data set of 5,503 instances of terrorists’ activities in Nigeria was obtained and a pattern recognition model was built using Artificial Neural Networks (ANN) with 70%, 15%, and 15% data splits for training, validation, and testing respectively. A 10-10-6 ANN architecture was designed and trained using the scaled conjugate gradient backpropagation algorithm. The training was carried out using Matlab’s neural network pattern recognition toolkit. In order to numerically represent categorical data, a sort-order scheme was developed by the authors and utilized. The results showed average percentage scores for accuracy, precision, recall and F1-score as 99.89, 99.96, 100 and 99.98 respectively. This showed acceptable performance. The developed model is therefore considered a robust one for recognition of terrorists’ patterns in Nigeria. This would assist security agencies to deal with terrorists’ incidences with high intelligent information and advanced preparation for prevention and control. The developed model is highly recommended for use by the security agencies in the country.

Analysis of PEM and AEM electrolysis by neural network pattern recognition, association rule mining and LIME

In this work, as an extension of previous machine learning studies, three novel techniques, namely local interpretable model-agnostic explanations (LIME), neural network pattern recognition and association rule mining (ARM) were utilized for proton exchange membrane (PEM) and anion exchange membrane (AEM) electrolyzer database for hydrogen production. The main goal of LIME was to determine the positive or negative effects of a variety of descriptor variables on current density, power density and polarization. Using this technique, it was possible to uncover rules or paths that lead to high current density, low power density and low polarization. ARM provided the dominant rules leading to high current density such as using ELAT as the cathode gas diffusion layer, using pure Pt on the cathode surface and using pure carbon as the cathode support. In addition, LIME and neural network pattern recognition successfully uncovered the importance of catalytic materials such as cathode/anode support/surface elements, operational variables like K2CO3 or KOH concentration in the electrolyte, certain membrane types, gas diffusion layers, and applied potential on current density. It was then concluded that machine learning can help determine the ideal conditions for developing a PEM and AEM electrolyzer to maximize hydrogen generation, which can also guide future research.

Crack Fault Diagnosis and Location Method for a Dual-Disk Hollow Shaft Rotor System Based on the Radial Basis Function Network and Pattern Recognition Neural Network

The crack fault is one of the most common faults in the rotor system, and researchers have paid close attention to its fault diagnosis. However, most studies focus on discussing the dynamic response characteristics caused by the crack rather than estimating the crack depth and position based on the obtained vibration signals. In this paper, a novel crack fault diagnosis and location method for a dual-disk hollow shaft rotor system based on the Radial basis function (RBF) network and Pattern recognition neural network (PRNN) is presented. Firstly, a rotor system model with a breathing crack suitable for a short-thick hollow shaft rotor is established based on the finite element method, where the crack’s periodic opening and closing pattern and different degrees of crack depth are considered. Then, the dynamic response is obtained by the harmonic balance method. By adjusting the crack parameters, the dynamic characteristics related to the crack depth and position are analyzed through the amplitude-frequency responses and waterfall plots. The analysis results show that the first critical speed, first subcritical speed, first critical speed amplitude, and super-harmonic resonance peak at the first subcritical speed can be utilized for the crack fault diagnosis. Based on this, the RBF network and PRNN are adopted to determine the depth and approximate location of the crack respectively by taking the above dynamic characteristics as input. Test results show that the proposed method has high fault diagnosis accuracy. This research proposes a crack detection method adequate for the hollow shaft rotor system, where the crack depth and position are both unknown.

Using artificial neural networks for pattern recognition of post-surgical infections

The objective is to use Artificial Intelligence (AI) for identifying which surgical patients have a likelihood ratio of developing an infection. We included in the study all the patients who underwent surgeries with wound class considered clean at a regional public hospital in Brazil. The first step was a retrospective analysis of risk factors and a correlation test for identifying which clinical variables are best related to post-discharge infections. Then, we developed an Artificial Neural Network (ANN) for pattern recognition to detect incidence of infections. The ANN can make accurate predictions in 77.3% of the cases in which an infection will occur, and the AUROC of the model is 0.9050. Thus, it is possible to take actions before the patients develop it, improving the quality of life and mental health as well as avoid increasing costs.

A SOM-CNN Algorithm for NLOS Signal Identification

Aiming at the problem of non-line of sight (NLOS) signal recognition for Ultra Wide Band (UWB) positioning, we utilize the concepts of Neural Network Clustering and Neural Network Pattern Recognition. We propose a classification algorithm based on self-organizing feature mapping (SOM) neural network batch processing, and a recognition algorithm based on convolutional neural network (CNN). By assigning different weights to learning, training and testing parts in the data set of UWB location signals with given known patterns, a strong NLOS signal recognizer is trained to minimize the recognition error rate. Finally, the proposed NLOS signal recognition algorithm is verified using data sets from real scenarios. The test results show that the proposed algorithm can solve the problem of UWB NLOS signal recognition under strong signal interference. The simulation results illustrate that the proposed algorithm is significantly more effective compared with other algorithms.

Machine Learning Methods for Diabetes Prevalence Classification in Saudi Arabia

Machine learning algorithms have been widely used in public health for predicting or diagnosing epidemiological chronic diseases, such as diabetes mellitus, which is classified as an epi-demic due to its high rates of global prevalence. Machine learning techniques are useful for the processes of description, prediction, and evaluation of various diseases, including diabetes. This study investigates the ability of different classification methods to classify diabetes prevalence rates and the predicted trends in the disease according to associated behavioural risk factors (smoking, obesity, and inactivity) in Saudi Arabia. Classification models for diabetes prevalence were developed using different machine learning algorithms, including linear discriminant (LD), support vector machine (SVM), K -nearest neighbour (KNN), and neural network pattern recognition (NPR). Four kernel functions of SVM and two types of KNN algorithms were used, namely linear SVM, Gaussian SVM, quadratic SVM, cubic SVM, fine KNN, and weighted KNN. The performance evaluation in terms of the accuracy of each developed model was determined, and the developed classifiers were compared using the Classification Learner App in MATLAB, according to prediction speed and training time. The experimental results on the predictive performance analysis of the classification models showed that weighted KNN performed well in the prediction of diabetes prevalence rate, with the highest average accuracy of 94.5% and less training time than the other classification methods, for both men and women datasets.

Investigation of Adversarial Attacks on Pattern Recognition Neural Networks

This article discusses the algorithm for creating a neural network based on pattern recognition. Several types of attacks on neural networks are considered, the main features of such attacks are described. An analysis of the Adversarial attack was carried out. The results of experimental testing of the proposed attack are presented. Confirmation of the hypothesis about the decrease in the accuracy of recognition of the neural network during the implementation of the attack by an attacker was obtained.

Pattern recognition in multi-synaptic photonic spiking neural networks based on a DFB-SA chip

Pattern recognition in multi-synaptic photonic spiking neural networks based on a DFB-SA chip

Developing an Efficient Cancer Detection and Prediction Tool Using Convolution Neural Network Integrated with Neural Pattern Recognition.

The application of computational approaches in medical science for diagnosis is made possible by the development in technical advancements connected to computer and biological sciences. The current cancer diagnosis system is becoming outmoded due to the new and rapid growth in cancer cases, and new, effective, and efficient methodologies are now required. Accurate cancer-type prediction is essential for cancer diagnosis and treatment. Understanding, diagnosing, and identifying the various types of cancer can be greatly aided by knowledge of the cancer genes. The Convolution Neural Network (CNN) and neural pattern recognition (NPR) approaches are used in this study paper to detect and predict the type of cancer. Different Convolution Neural Networks (CNNs) have been proposed by various researchers up to this point. Each model concentrated on a certain set of parameters to simulate the expression of genes. We have developed a novel CNN-NPR architecture that predicts cancer type while accounting for the tissue of origin using high-dimensional gene expression inputs. The 5000-person sample of the 1-D CNN integrated with NPR is trained and tested on the gene profile, mapping with various cancer kinds. The proposed model's accuracy of 94% suggests that the suggested combination may be useful for long-term cancer diagnosis and detection. Fewer parameters are required for the suggested model to be efficiently trained before prediction.

Natural Language Processing to Extract Information from Portuguese-Language Medical Records

Studies that use medical records are often impeded due to the information presented in narrative fields. However, recent studies have used artificial intelligence to extract and process secondary health data from electronic medical records. The aim of this study was to develop a neural network that uses data from unstructured medical records to capture information regarding symptoms, diagnoses, medications, conditions, exams, and treatment. Data from 30,000 medical records of patients hospitalized in the Clinical Hospital of the Botucatu Medical School (HCFMB), São Paulo, Brazil, were obtained, creating a corpus with 1200 clinical texts. A natural language algorithm for text extraction and convolutional neural networks for pattern recognition were used to evaluate the model with goodness-of-fit indices. The results showed good accuracy, considering the complexity of the model, with an F-score of 63.9% and a precision of 72.7%. The patient condition class reached a precision of 90.3% and the medication class reached 87.5%. The proposed neural network will facilitate the detection of relationships between diseases and symptoms and prevalence and incidence, in addition to detecting the identification of clinical conditions, disease evolution, and the effects of prescribed medications.

Very Short-Term Forecast: Different Classification Methods of the Whole Sky Camera Images for Sudden PV Power Variations Detection

Solar radiation is by nature intermittent and influenced by many factors such as latitude, season and atmospheric conditions. As a consequence, the growing penetration of Photovoltaic (PV) systems into the electricity network implies significant problems of stability, reliability and scheduling of power grid operation. Concerning the very short-term PV power production, the power fluctuations are primarily related to the interaction between solar irradiance and cloud cover. In small-scale systems such as microgrids, the adoption of a forecasting tool is a brilliant solution to minimize PV power curtailment and limit the installed energy storage capacity. In the present work, two different nowcasting methods are applied to classify the solar attenuation due to clouds presence on five different forecast horizons, from 1 to 5 min: a Pattern Recognition Neural Network and a Random Forest model. The proposed methods are tested and compared on a real case study: available data consists of historical irradiance measurements and infrared sky images collected in a real PV facility, the SolarTechLAB in Politecnico di Milano. The classification output is a range of values corresponding to the future value assumed by the Clear Sky Index (CSI), an indicator allowing to account for irradiance variations only related to clouds passage, neglecting diurnal and seasonal influences. The developed models present similar performance in all the considered time horizons, reliably detecting the CSI drops caused by incoming overcast and partially cloudy sky conditions.