Real-time Architecture Research Articles

Closed-Loop Fuzzy Energy Regulation in Patients With Hypercortisolism via Inhibitory and Excitatory Intermittent Actuation

Hypercortisolism or Cushing's disease, which corresponds to the excessive levels of cortisol hormone, is associated with tiredness and fatigue during the day and disturbed sleep at night. Our goal is to employ a wearable brain machine interface architecture to regulate one's energy levels in hypercortisolism. In the present simulation study, we generate multi-day cortisol profile data for ten subjects both in healthy and disease conditions. To relate an internal hidden cognitive energy state to one's cortisol secretion patterns, we employ a state-space model. Particularly, we consider circadian upper and lower bound envelopes on cortisol levels, and timings of hypothalamic pulsatile activity underlying cortisol secretions as continuous and binary observations, respectively. To estimate the hidden cognitive energy-related state, we use Bayesian filtering. In our proposed architecture, we infer one's cognitive energy-related state using wearable devices rather than monitoring the brain activity directly and close the loop utilizing fuzzy control. To model actuation in the real-time closed-loop architecture, we simulate two types of medications that result in increasing and decreasing the energy levels in the body. Finally, we close the loop using a knowledge-based control approach. The results on ten simulated profiles verify how the proposed architecture is able to track the energy state and regulate it using hypothetical medications. In a simulation study based on experimental data, we illustrate the feasibility of designing a wearable brain machine interface architecture for energy regulation in hypercortisolism. This simulation study is a first step toward the ultimate goal of managing hypercortisolism in real-world situations.

Application of Machine Learning for Emotion Classification

In this paper we propose an implement a general convolutional neural network (CNN) building framework for designing real-time CNNs. We validate our models by creat- ing a real-time vision system which accomplishes the tasks of face detection, gender classification and emotion classification simultaneously in one blended step using our proposed CNN architecture. After presenting the details of the training pro- cedure setup we proceed to evaluate on standard benchmark sets. We report accuracies of 96% in the IMDB gender dataset and 66% in the FER-2013 emotion dataset. Along with this we also introduced the very recent real-time enabled guided back- propagation visualization technique. Guided back-propagation uncovers the dynamics of the weight changes and evaluates the learned features. We argue that the careful implementation of modern CNN architectures, the use of the current regu- larization methods and the visualization of previously hidden features are necessary in order to reduce the gap between slow performances and real-time architectures. Our system has been validated by its deployment on a Care-O-bot 3 robot used during RoboCup@Home competitions. All our code, demos and pre- trained architectures have been released under an open-source license in our public repository.

A Real-Time Architecture for Pruning the Effectual Computations in Deep Neural Networks

Integrating Deep Neural Networks (DNNs) into the Internet of Thing (IoT) devices could result in the emergence of complex sensing and recognition tasks that support a new era of human interactions with surrounding environments. However, DNNs are power-hungry, performing billions of computations in terms of one inference. Spatial DNN accelerators in principle can support computation-pruning techniques compared to other common architectures such as systolic arrays. Energy-efficient DNN accelerators skip bit-wise or word-wise sparsity in the input feature maps (ifmaps) and filter weights which means ineffectual computations are skipped. However, there is still room for pruning the effectual computations without reducing the accuracy of DNNs. In this paper, we propose a novel real-time architecture and dataflow by decomposing multiplications down to the bit level and pruning identical computations in spatial designs while running benchmark networks. The proposed architecture prunes identical computations by identifying identical bit values available in both ifmaps and filter weights without changing the accuracy of benchmark networks. When compared to the reference design, our proposed design achieves an average per layer speedup of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\times 1.4$ </tex-math></inline-formula> and an energy efficiency of <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\times 1.21$ </tex-math></inline-formula> per inference while maintaining the accuracy of benchmark networks.

A Real-Time Multi-Stage Architecture for Pose Estimation of Zebrafish Head with Convolutional Neural Networks

In order to conduct optical neurophysiology experiments on a freely swimming zebrafish, it is essential to quantify the zebrafish head to determine exact lighting positions. To efficiently quantify a zebrafish head's behaviors with limited resources, we propose a real-time multi-stage architecture based on convolutional neural networks for pose estimation of the zebrafish head on CPUs. Each stage is implemented with a small neural network. Specifically, a light-weight object detector named Micro-YOLO is used to detect a coarse region of the zebrafish head in the first stage. In the second stage, a tiny bounding box refinement network is devised to produce a high-quality bounding box around the zebrafish head. Finally, a small pose estimation network named tiny-hourglass is designed to detect keypoints in the zebrafish head. The experimental results show that using Micro-YOLO combined with RegressNet to predict the zebrafish head region is not only more accurate but also much faster than Faster R-CNN which is the representative of two-stage detectors. Compared with DeepLabCut, a state-of-the-art method to estimate poses for user-defined body parts, our multi-stage architecture can achieve a higher accuracy, and runs 19x faster than it on CPUs.

Instrumental Non-Invasive Diagnostics of the Depth of Skin Burn:&#x0D; Current Opportunities and Unsolved Problems

Introduction. Burns of various etiologies are on the 3-4 place in frequency among all injuries, they are associated with a high percentage of complications, disability and mortality. Currently, a single algorithm for objective verification of the depth of skin damage has not been developed. Numerous diagnostic technologies used to solve this problem require systemic analysis.The aim of study was to analyze current literature data on technologies for instrumental diagnostics of the depth of thermal damage to the skin.Materials and methods. A search for literary sources in databases was conducted using eLIBRARY.RU, PubMed, Cyberleninka databases. The period of publications was limited to 2011-2020.Results. The method of objective "gold standard" verification of thermal damage to the skin remains a biopsy followed by morphological examination, but the method's traumaticity does not allow it to be used for monitoring the course of the process in dynamics, for multi-focal studies. Among non-invasive methods, dermatoscopy and video microscopy are noted, but the small imaging area and the need for direct contact of the device with the skin surface limit the use of these methods. Laser Doppler flowmetry and imaging are fairly accurate non-contact methods that allow assessing the condition of a burn wound in real time. The method is not applicable to wounds with blisters, insufficient sanitation. Laser speckle contrast imaging is used to evaluate blood flow and is based on the analysis of speckle structure fluctuations. The method allows determining the degree of burn damage depending on changes in the blood flow. However, the method is difficult to apply with a vessel diameter sized 40 microns. The method is sensitive to movement and internal factors. Infrared and Raman spectroscopy are technique for evaluating vibrational patterns in a particular spectrum, these can be used to identify a molecule or determine its structure. Simultaneous use of spatial frequency visualization and speckle imaging demonstrated a high level of correlation with the results of histological research, which makes the use of these methods promising. Ultrasound also provides a proper correlation with histological data, but it is an imperfect method due to inaccuracies in determining the thickness of the epidermis, dermis, and scar condition. Optical coherence tomography (OCT) is a non-invasive optical method that allows obtaining high-resolution images of skin architecture in real time. The high level of compliance of the visualized structures, in particular, layers, appendages and vessels of the skin, with histological findings, gave this method a name "optical biopsy". Anatomical congruence of normal and damaged skin was established in an experiment between histology and OCT.Conclusion. Early assessment of the depth of thermal damage to the skin is crucial for choosing a personalized treatment strategy for the burned. None of the modern diagnostic methods is universal. Multimodal approaches to diagnostics are the most effective.

RETRACTED ARTICLE: Pedestrian identification using motion-controlled deep neural network in real-time visual surveillance

In the computer vision applications such as security surveillance and robotics, pedestrian identification shows much attention in the last decade. This is usually achieved by human biometrics. Besides human biometrics, sometimes it is required to identify pedestrians at a distance. This could be accomplished based on a fact of different whole-body appearances. The real-time pedestrian identification is a challenging task due to several factors such as illumination effects, noise, change in viewpoint, and video resolution. The more recent, the deep neural network (DNN) shows a massive performance for various real-world applications. In this article, we present a real-time architecture for pedestrian identification using motion-controlled DNN. In the proposed architecture, the motion vectors are calculating using optical flow and then utilized in the next step, named features extraction. Two types of features, such as HOG and DNN, are computing. The pre-trained VGG19 CNN model is employing and trained through transfer learning. The deep learning features are extracted from two layers—fully connected layers 7 and 8. Also, we proposed a feature selection method named Bayesian modeling along with LSVM. The best selected features of both HOG and DNN are finally fused in one matrix for final identification. The multi-class support vector machine classifier is used for final identification. The videos are recording in the real-time environment for the experimental process and achieve an average accuracy of 98.62%. Overall, identification accuracy shows the effectiveness of the proposed approach.

Experimental Implementation and Validation of a Broadband LTI Energy-Maximizing Control Strategy for the Wavestar Device

This study addresses the experimental validation of a linear time-invariant (LTI) energy-maximizing control strategy for wave energy converters (WECs), applied to a 1/20 scale Wavestar WEC. To fulfill this objective, system identification routines are utilized to compute a mathematical (parametric) model of the input–output dynamics of the device, suitable for control design and implementation. With this parametric model, the so-called LiTe-Con energy-maximizing strategy, recently published in the literature, is designed, synthesized, and tested under irregular wave excitation in the wave basin at Aalborg University. Given that the LiTe-Con requires instantaneous knowledge of the wave excitation effects, estimates are provided by means of an unknown-input Kalman filter, designed in close synergy with the so-called internal model principle. For the experimental assessment, both controller and estimator are directly implemented in a real-time architecture. The performance of the LiTe-Con is evaluated in terms of energy-absorption, showing consistent results with respect to those obtained in numerical simulation, hence validating the LiTe-Con controller in a realistic real-time scenario.

RETRACTED: Deep learning for real-time semantic segmentation: Application in ultrasound imaging

A real-time architecture of medical image semantic segmentation called Fully Convolution dense Dilated Network, is proposed to improve the segmentation efficiency while ensuring high accuracy. Considering low resolution and contrast, interferences of shadows, as well as differences in nodules’ position and size, accurate ultrasound images’ segmentation cannot be obtained easily. Therefore, a novel layer that integrates the advantages of dense connectivity, dilated convolutions and factorized filters, is proposed in an attempt to remain efficient while retaining remarkable accuracy. Dense connectivity combines low-level fine segmentation with high-level coarse segmentation to extract more features from ultrasound images. Dilated convolution can expand the receptive field of the filter, and the problem of differences in nodules’ size and position can be solved with different sizes of filters. This study also introduces factorized filters into the network to further optimize the efficiency of the model. In addition, aiming at the class imbalance problem in medical image semantic segmentation, a loss function optimization method is proposed which further improves the accuracy of the network. A thorough set of experiments based on thyroid dataset show that the proposed model achieves state-of-the-art performance in terms of robustness and efficiency.

Hardware-software implementation of the compression and transmission of seismic data

A real-time wireless network architecture for seismic data ac-quisition system based on multi-level radio network is proposed. The sin-gle-chip programmable transceiver Si446x that allows creating the first level of the network is considered. Physical layer with GFSK modulation in the sub-gigahertz band is considered. Data link layer implementation with generic frame structure for a network, operating in synchronous mode, is presented. An algorithm for seismic data accumulating and com-pressing is presented. The comparison of existing and proposed solutions is given.

Erratum to "A Low Power and Real-Time Architecture for Hough Transform Processing Integration in a Full HD-Wireless Capsule Endoscopy".

Presents corrections to author information in the above named paper.

Art-based Public Sociology

Dieser Beitrag reflektiert eine zweisemestrige Lehrforschung im Fachbereich Soziologie an der Friedrich-Schiller-Universität Jena. Erforscht wurde mittels einer teilnehmenden Beobachtung der Prozess des ersten internationalen Echtzeit-Architekturfestivals »72 Hour Urban Action«, das vom 2. bis 5. Mai 2019 inmitten der Großwohnsiedlung Lobeda-West in der Peripherie von Jena stattfand. Die Mission der architektonischen Intervention bestand darin, zehn neue Orte der Begegnung zu schaffen und die Anwohner*innen des Quartiers für die Gestaltung des öffentlichen Raumes zu aktivieren. Als Feldforscher*innen arbeiteten wir eng mit Stadtplaner*innen, Kulturschaffenden und Künstler*innen zusammen und wurden zu einem Teil des Untersuchungsgegenstandes. Wir entwickelten künstlerische Präsentationsformen und trugen relevantes Wissen über den Stadtteil zur Realisierung des Architekturfestivals bei. Welche Potenziale eine partizipative Stadtraumentwicklung und welche Rolle performative Sozialwissenschaft dabei spielen kann, soll in diesem Beitrag ausgeleuchtet werden.

Automatic waste detection by deep learning and disposal system design

As Dubai aims to become a greener city by the year 2021, efforts are currently being made by the government to devise a more efficient and innovative approach to tackling solid-waste-management issues in the city. With a much higher rate of recycling of trash, there arises a need to find a better approach to classifying this trash with increased efficiency. Machine learning techniques can be employed to classify trash into different recycling categories so that it is easier to recycle waste. In this paper, an automatic waste-classification system is proposed using a deep learning algorithm to classify waste as metal, paper, plastic and non-recyclable waste. The classification was performed through this computer vision approach by using the AlexNet convolutional neural network architecture in real time so that the waste can be dropped into the appropriate chambers as soon as it is thrown into dustbins. The data set used to train the system consisted of images collected from the Internet, as well as hand-collected images. The model used was tested for classification of different types of trash and was found to show a high accuracy, as discussed in the result section.

Wafer sojourn time fluctuation analysis for time-constrained dual-arm multi-cluster tools with activity time variation

ABSTRACT In semiconductor manufacturing systems, a time-constrained multi-cluster tool should be scheduled such that a wafer stays in a process chamber in a given time range to satisfy a wafer residency time constraint. In practice, activity time is subject to variation. It could lead to some fluctuation of wafer residency time in a process chamber. Hence, it is crucial to analyze how wafer residency time varies with activity time variation. This issue is especially challenging for multi-cluster tools. This work focuses on determining the exact upper bound of wafer sojourn time delay resulted from activity time variation for dual-arm multi-cluster tools. After discussing their dynamic behaviours, it presents a two-level real-time operational architecture and a real-time control policy. Based on them, this work derives for the first time an efficient algorithm to calculate the exact upper bound of wafer sojourn time delay in a process chamber. As a result, engineers can test whether a given schedule is feasible. Several examples of industrial significance are used to demonstrate the application of the proposed approach.

Agent-based dynamic scheduling model for product-driven production

Goal: This research provides specific solution for dynamic scheduling of product-driven production with unique level of detail and original architecture.&#x0D; Design / Methodology / Approach: Design process of scheduling problem-solving MAS is divided into three steps: agent encapsulation of entities participating in scheduling, including concept of agents and responsibilities they assume, system architecture and topology of the agents network, detailed design of decision scheme of individual agents.&#x0D; Results: Production processes take place in dynamic environment and have to react to numerous real-time events, hence reschedule the production by a new design and implement agent-based model in order to solve dynamic flexible job shop scheduling problem in product-driven production environment.&#x0D; Limitations of the investigation: Designed model counts with simple agents behaving on condition-action rules. These agents could be replaced by more sophisticated types of agents such as utility-based or learning agents. Also the implemented coordination mechanism ensuring global view on the scheduling problem is rather simple.&#x0D; Practical implications: Via simulations of realistic production scenarios it was expected to prove an applicability of specific solution of how bringing the intelligence to the lower levels of control system may be used in dynamic scheduling. Based on elaboration of theoretical basis, principles were identified as suitable or widely used in design of such model.&#x0D; Originality / Value: This research provides specific real-time architecture for a multi-agents dynamic scheduling of product-driven production with unique level of detail and scenarios analysis.

VLSI Architecture of Cubic SP line Interpolation on FPGA

A novel time frequency analysis method was proposed by N.E.Huang known as Hilbert Huang Transform which, can be used for analyzing and processing real world signals. The Intrinsic Mode Functions (IMF) is the key part of this algorithm, in this part the empirically decomposed signal data points uses the cubic spline interpolation for connecting maximum and minimum points to connect lower and upper envelope of the processed signal. This paper presents the real time architecture for hardware implementation of natural cubic spline interpolation. The architecture of proposed cubic spline is using the properties of continuous cubic and linear polynomials. The experimental results showed that our proposed architecture gets better result than previous proposal implemented on Spartan 6 based FPGA board.

Real time 3-D ultrasound diagnostic imaging system including 3-D adaptive beamforming processing

This paper describes the analog front end module and the computing architecture components of a fully digital real-time 3-D ultrasound system. The computing architecture is designed to allow for an efficient implementation of a 3-D adaptive beamformer that has the capability to improve the angular image resolution of a planar array by approximately four times. The complex 3-D beamforming structure is decomposed into two steps of line array beamformers and this kind of decomposition process for the 3-D beamformer allows for its efficient implementation into the highly parallelized multi-processor based computing architecture for real time 3-D ultrasound imaging applications providing 20 volumes per second at a full opening angle of 80 deg (azimuth and elevation). The main objective of this paper is to describe the details of the system processing requirements and design consideration for the computing architecture and provide the experimental results showing that the proposed implementation can achieve the targeted frame rate. An easy to use user interface in combination with a decision-support process provides the possibility for a rapid and automated diagnosis of internal injuries like bleeding or facilitates image guided surgery.

FPGA Based Architecture Implementation for Epileptic Seizure Detection Using One Way ANOVA and Genetic Algorithm.

Epilepsy is a brain disorder which produces recurrent seizures as a storm of the electrical activity of the brain. 70 millions of people living with epilepsy in the world and most of them are from developing countries and near about 12 millions of people are residing from India. In rural areas, seizure disorder is not treated seriously so there is a need for awareness and availability of proper medication. Recurring seizures are the major source of diagnosis of epilepsy so real-time prediction using analytical methods is a need of the research in this area. Electroencephalographic (EEG) signals are the rich source of the early diagnosis of epilepsy. The basic objective of the work is to proposed real time architecture which could be included in existing EEG monitoring and measuring instruments to mark the seizure occurrence. This will facilitate medical practitioners monitoring primary status of patients and understanding frequency of seizure occurrence. Thus the proposed work provide real-time architecture or improved performance reconfigurable solution to contribute in designing real-time seizure detection system. The EEG processing architecture is designed and implemented in this work, which will add values to the existing EEG monitoring and recording system.

Improving Imaging Quality of Real-time Fourier Single-pixel Imaging via Deep Learning.

Fourier single pixel imaging (FSPI) is well known for reconstructing high quality images but only at the cost of long imaging time. For real-time applications, FSPI relies on under-sampled reconstructions, failing to provide high quality images. In order to improve imaging quality of real-time FSPI, a fast image reconstruction framework based on deep learning (DL) is proposed. More specifically, a deep convolutional autoencoder network with symmetric skip connection architecture for real time 96 × 96 imaging at very low sampling rates (5–8%) is employed. The network is trained on a large image set and is able to reconstruct diverse images unseen during training. The promising experimental results show that the proposed FSPI coupled with DL (termed DL-FSPI) outperforms conventional FSPI in terms of image quality at very low sampling rates.

A new beamforming method and hardware architecture for real time two way dynamic depth focusing

The Total Focusing Method (TFM) yields a focused image in emission and in reception while Phased Array (PA) imaging provides Dynamic Depth Focusing (DDF) in reception only. Besides, most NDE applications have two propagation media, where refraction at the interface complicates time-of-flight (TOF) and focal law computations. This affects especially TFM, which must compute the TOFs from all elements to image pixels and use them to select the data for imaging.A new method with real-time Dynamic Depth Full Focusing (DDFF), in emission and reception, is proposed in this work. It is called Total Focusing Phased Array (TFPA) because it uses concepts of TFM and PA. Omnidirectional emissions are used to create a synthetic aperture as in TFM, while beamforming is carried out along scan lines as in PA, simplifying the delay calculation in the presence of interfaces and providing an efficient hardware implementation.Refraction at the interface between two media is eliminated by a Virtual Array (VA) that converts such scenario into a simple homogeneous medium. Propagation can be considered along scan lines from the virtual array at constant speed, as in homogeneous media. Strict dynamic focusing is performed in real-time, an important difference with other approaches that require iterative Fermat search to get the focal laws for every imaged point. With TFPA only 3 parameters per element and scan line are required to perform this task.Experiments are carried out to compare the three techniques, PA, TFM and TFPA. TFM and TFPA yield similar image quality, offering improved depth of field and resolution over PA. On the other hand, TFPA avoids most of the burden for computing TOFs and operates in real time with one or two media propagation.

Design and implementation of MIMO radar real‐time imaging system based on multicore DSP

Multiple-input multiple-output (MIMO) imaging radar is a new kind of new imaging radar in recent years. It can achieve faster data acquisition speed than the traditional ground-based SAR due to the ability to get a large number of multi-channel data in a short period of time. However, traditional single-core digital signal processors (DSPs) have limited space and ability to deal with such large amount of data. To solve this problem, a real-time image processing software architecture for MIMO imaging radar algorithm is designed and carried out based on TMS320C6678 DSP of Texas Instruments (TI), which is the highest-performance fixed/floating-point DSP based on TI's KeyStone multicore architecture. The efficient imaging algorithms mapping from Matlab to DSP is realised, with efficient task allocation, resource reuse of echo data, and processing variable, and using the method of multicore parallel processing.