Underwater wireless acoustic sensor networks (UWASNs) have emerged as a powerful communication technology for discovering and extracting data in aquatic environments. UWASNs have numerous applications in areas such as fisheries, resource exploration, mine reconnaissance, oil and gas inspection, marine exploration and military surveillance. However, these applications are limited by the capacity of networks to detect, discover, transmit, and forward big data. In particular, transmitting and receiving large volumes of data requires great lengths of time and substantial power, and thus fails to meet the real-time constraints. This problem has motivated us to focus on developing an underwater computer-embedded system capable of efficient big-data management. Thus, we have developed methods to discover and extract valuable information beneath the ocean using data-mining approaches. Previously, we introduced real-time underwater system architectures (RTUSAs) that use a single computer. In this study, we extend our results and propose a new RTUSA for large-scale networks. This novel RTUSA uses multi-computers and aims to enhance the reliability of our proposed system. Determining the optimal location of computers with respect to their membership of acoustic sensor nodes, so as to minimize delay time, power consumption, and balance loads, are NP-hard problems. Therefore, we propose a heuristic approach that enables optimization of computer locations and their memberships of acoustic sensor nodes. We conduct simulations to show the merits of our findings and measure the performance of our proposed solution.
Underwater Wireless Sensor Networks(UWSNs) have emerged as a promising technology that is used to monitor underwater environment. Applications of UWSNs are numerous such as oil and gas pipeline monitoring, underwater animal detection, and object of interest detection. Automated Underwater Vehicles (AUVs) have been used to monitor underwater environment [13]. One of the significant challenges of AU...
In this paper, a derivative-based MUSIC (DB-MUSIC) algorithm for two-dimensional (2-D) direction-of-arrival (DOA) estimation is proposed using an L-shaped uniform array. It transforms the traditional 2-D search problem into a one-dimensional (1-D) one using a derivative based optimization method, taking into consideration the structure of the steering vector and the associated cost function. As a result, the proposed algorithm has a significantly low computational complexity with the additional benefit of no need for 2-D angle pairing. Simulation results show that the proposed algorithm has better estimation accuracy than some existing representative 2-D DOA estimation algorithms falling into the same category, i.e., low complexity through 1-D search with no need for pairing.
Nowadays, fires in forest areas are very frequent, mainly caused by climate change and bad practices by the people who live in these areas. In the world the climatic "El Niño" phenomenon has intensified in recent years, increasing the frequency of forest fires, due to high temperatures and prolonged periods of drought that occur. Most forest fires are detected visually and from the ground or from the air using a helicopter; this method is not very efficient since it takes too long to alert the relief corps and requires well-organized logistics. The lack of early detection means has been evident in the events that have occurred in recent months (last fires) and it can be concluded that there are not enough measures to counteract this problem.The purpose of this article is to evaluate the performance of different CNN models pre-trained in the classification of forest fire images, which can be applied in economic development cards such as a Raspberry.
Value Iteration (VI) is a powerful, though time consuming, approach to solve reinforcement learning problems modeled as Markov Decision Processes (MDPs). In this paper, we explore strategies to run the sate-of-the-art cache efficient algorithm for VI developed by us [1], [2] on a multicore processor. We demonstrate a speedup of up to 2.59 on a 10-core multiprocessor using 20 threads on popular benchmark data. The speedup for the parallelized portion of the computation is up to 5.89.
Attention mechanism has been applied to the weakly supervised sound event detection (SED) and has achieved state-of-the-art performance, but most methods only concentrate along the time axis. In this paper, we propose the multi-scale time-frequency attention (MTFA) method to capture the intrinsic features at different scales both in time and frequency domain for audio tagging (AT) and SED. Our model is a unified network which can perform AT and SED simultaneously, it produces multi-scale attention-aware representations for SED with MTFA module, and a global pooling module maps the representations to presence probability of corresponding audio event for AT. To evaluate the proposed method, we conduct experiments on Task4 of Detection and Classification of Acoustic Scenes and Events (DCASE) challenge, and it achieves 57.9% (F1-score) in AT task and 0.71 (error rate) in SED task on evaluation set, which is comparable to the state-of-the-art results in the challenge.
Cover song identification (CSI) is a challenging task in the music information retrieval (MIR) community.The employment of convolutional neural networks (CNN) have significantly improved the performance of CSI systems, especially CNN designed to be invariant against key transpositions. In this paper, we propose MulKINet, a multi-stage CNN architecture that preserve the property of key invariance while its representational ability is substantially enhanced. Combined with three options for building blocks, channel and temporal attention mechanism, we present an accurate and fast CSI system.
Spatial audio is one of the most essential parts of immersive audio-visual experience such as virtual reality (VR), which reproduces the inherent spatiality of sound and the correspondence of audio-visual experience. Ambisonics is the dominant spatial audio solution due to its flexibility and fidelity. However, the production of Ambisonics audio is difficult for the public because of the requirements of expensive equipments or professional music production ability. In this work, an end-to-end Ambisonics generator for panorama video is proposed. To improve the perception of directional sound, we assume that sound field is composed of a primary sound source and an ambient sound without spatiality, and a Temporal Convolutional Network (TCN) based Primary Ambient Extractor (PAE) is proposed to separate the two parts of sound field. The directional sound is spatially encoded by the weights from audio-visual fusion network added by ambient part. Our network is evaluated with panorama video clips with first order Ambisonics. The results show that the proposed approach outperforms other methods in terms of objective evaluations.
This paper presents a Multilayer Perceptron and Support Vector Machine algorithms approach to predict the number of COVID19 infections in different countries of America. It intends to serve as a tool for decision-making and tackling the pandemic that the world is currently facing. The models were trained and tested using open data from the European Union repository where a time series of confirmed contagious cases was modeled until May 25, 2020. The hyperparameters as number of neurons per layer were set up using a tabu list algorithm. The countries selected to carry out the study were Brazil, Chile, Colombia, Mexico, Peru and the United States. The metrics used are Pearson's correlation coefficient (CP), Mean Absolute Error (MAE), and Mean Percentage Error (MPE). For the testing stage we obtained the following results: Brazil, CP=0.65, MAE=2508 and MPE=17%; Chile, CP=0.64, MAE=504, MPE=16%; Colombia, CP=0.83, MAE=76, MPE=9%; Mexico, CP=0.77, MAE=231, MPE=9%; Peru, CP=0.76, MAE=686, MPE=18% and the United States of America, CP=0.93, MAE=799, MPE=4%. This resulted in powerful machine learning tools although it is necessary to use specific algorithms depending on the data and the stage of the country’s pandemic.
Fall detection has been an important consideration in the field of human activity recognition and has garnered significant interest from researchers. A typical aim within fall detection systems is the determination of whether a fall has occurred or not. However, less attention has been provided to the problem of fall direction detection and severity. In this paper, we experiment with the detection of direction and severity in falls using the SisFall dataset. We perform this by using a combination of time and frequency domain features on inertial measurement sensor values along with a Support Vector Machine classifier. We are able to achieve promising results for the considered task.
Consensus boost and opinion guidance are two important problems during the opinion management process. Considering that opinion interaction with opinion dynamics, this paper formalizes the two problems as markov decision process. To solve the two problems with minimum cost, we proposes consensus boost algorithm and opinion guidance algorithm based on reinforcement learning. Meantime, we construct opinion management framework by combining consensus boost algorithm and opinion guidance algorithm which is beneficial to the opinion management of managers. Finally, through experimental analysis, we verify the effectiveness and properties of the proposed framework.
The study of heartbeats in electrocardiogram (ECG) signals is very important to sustain good health. Any anomalies in the heart rhythm can be detected by carefully studying the ECG signal. The detection of the QRS is obstructed by external and internal sources of noise. Automatic detection of the QRS is achieved by diminishing these noises to a minimum by different types of filtering such as band-pass filtering, wavelet transform, and applying thresholds. This paper presents a new method of QRS detection using discrete wavelet transform (DWT), median filtering, and adaptive multilevel thresholding (AMT). The proposed method is tested for the MIT-BIH Arrhythmia database and shows a high sensitivity of 99.74%, positive predictivity of 99.88%, and a detection error rate of 0.38%. In addition to this, the proposed technique is quite robust and can adapt to signals with a low signal-to-noise ratio.
In recent years, the demand for wrist wearable devices to monitor continuously critical physiological parameters in real time that are limited by designated hospital monitoring equipment is steadily increasing. In the medical field, one of the main issues that wearable devices could sufficiently address is the pervasive monitoring of vital signs and the corresponding health status assessment of the rapidly growing elderly population in real time. Main advantages in the adoption of wearable devices for the real time monitoring are the significant decrease of the cost both for the health system and subsequently the patient as well as the dramatic decrease of the waiting time in the hospital emergency rooms.Reflectance pulse oximetry being the right mode to be used at the wrist for measurements such as Heart Rate (HR), Peripheral Capillary Oxygen Saturation (SpO2) and Respiratory Rate (RR) imposes many technical challenges with its excessive sensitivity to all types of entailed artifacts due to arm/hand/body motions to be amongst the major ones.This work introduces a low-power wrist wearable device comprising a Photoplethysmography (PPG) array sensor special extraction algorithms to estimate HR and SpO2 parameters and a Multiple Linear Regression model, which after training performs considerable reduction of the imposed Motion Artifacts (Mas) thus enabling more accurate reading outputs.
Cloud Computing offers resources as a utility that can be accessed and rented via web browsers. It has also made it easy for paying for resources with different ways of pricing. However, that could result in overpaying for resources or underutilizing the reserved resources. In this paper, we focus on two main options in the pricing model, namely, pay-per-use and reserved instances. In the first one, users can pay for what they use only while the second option offers up to 72% discount but they have to pay in advance for the whole reserved period. In this paper, we present two algorithms for provisioning cloud resources to help cloud customers pick the most cost-effective plans for their jobs
Underwater wireless acoustic sensor networks (UWASNs) have been used as an efficient means of communication to discover and extract data in aquatic environments. Applications of UWASNs include marine exploration, mine reconnaissance, oil and gas inspection, marine exploration, and border surveillance and military applications. However, these applications are limited by the huge volumes of data involved in detection, discovery, transmission, and forwarding. In particular, the transmission and receipt of large volumes of data require an exhaustive amount of time and substantial power to execute, and may still fail to meet real-time constraints. This shortcoming directed our research focus to the advancement of an underwater computer embedded system to meet the required limitations. Our research activities have included the extraction of valuable information from under the ocean using data mining approaches. We previously introduced real-time underwater system architectures that use a single computer. In this study, we extend our results and propose a new real-time underwater system architecture for large-scale networks. This architecture uses multiple computers to enhance its reliability. Determining the optimal locations of computers and their membership of acoustic sensors with minimum delay time, power consumption, and load balance is an NP-hard problem. We therefore propose a heuristic approach to find the optimal locations of computers and their membership of acoustic sensor nodes. We then develop sensor network topologies that reduce data-aggregation latency and data loss and increase the network lifespan. This paper merges heuristic solutions and topologies to achieve the best network performance. A simulation is performed to show the merit of our results and to measure the performance of our proposed solution.
Wavelet-feature Markov clustering algorithm for the remotely sensed data is based on an accurate description of abrupt spectral features and an optimized Markov clustering in the wavelet feather space. The peak points can be captured and identified by applying wavelet transform on the expanded multispectral data. The correlation ratio between the two samples is a statistical calculation of the matched peak point positions on the wavelet-feature within an adjustable spectrum domain or a range of wavelet scales. The evenly sampled data can be used to create class centers, depending on the correlation ratio threshold at each Markov step, accelerating the clustering speed by avoiding computation of Euclidean distance for traditional clustering algorithms, such as K-means and ISODATA. By applying a simulated annealing method and gradually shrunk clustering size, Markov clustering leads to the best class centers quickly at each clustering temperature. The experimental results about TM data have verified its acceptable clustering accuracy and high convergence velocity.
Recently, a variable frame rate imaging method based on Fourier transformation has been developed to increase resolution and reduce sidelobe. Experiments with the imaging methods including D&S, 1-angle HFR (HFR 1), 11-angle HFR (HFR 11), 19-angle HFR (HFR 19), and 91-angle HFR (HFR 91) have also been carried out. In the experiment, one linear array was used to construct 2D B-mode images for a tissue-equivalent phantom and pointer scatterer. The array had a center frequency of 2.5MHz, dimensions of 19.2mm×14mm, and 128 elements. The experiments on the resolution and sidelobe were done with pointer scatterer in the water tank. Results show that HFR 11, HFR 19, and HFR 91 have higher resolution than D&S at all depths. The sidelobe for HFR 1, HFR 11, HFR 19, D&S, and HFR 91 decreases in turn, and HFR 91 has the lowest sidelobe. The experiments on the contrast comparison between HFR and D&S method are made on one tissue-equivalent phantom, eight cones with different contrasts (−15dB, −10dB, −5dB, −2dB, 2dB, 4dB, 7.5dB and 12dB) over background. The contrast curves of eight cones for HFR 1, HFR 11, HFR 19, D&S, and HFR 91 shift downward in turn, which is compatible with their sidelobe property. The contrast recognition accuracy of HFR 91 is the best. All evaluation standards show that the high frame rate imaging method is better than the conventional delay and sum method if their frame rates are the same, so high resolution and low-sidelobe images can be constructed at a high frame rate with this Fourier method.
This paper presents a Quantum versus classical implemented of Machine learning (ML) algorithm applied to a diabetes dataset. Diabetes is a Sixth deadliest disease in the world and approximately 10 million new cases are registered every year worldwide. Using novel Quantum computing (QC) along with Quantum Machine Learning (QML) techniques in the healthcare system to improve and accelerate the computing of existing ML models that allows the different approach to understanding the complex patterns of the disease. The proposed system tackles a binary classification problem of patients with diabetes into two different classes: diabetes patients with acute diseases and diabetes patients without acute diseases. Our study compares classical and quantum algorithms, namely Decision Tree, Random Forest, Extreme Boosting Gradient and Adaboost, Qboost, Voting Model 1, Voting Model 2, Qboost Plus, New model 1 and New Model 2 along with an ensemble method which creates a strong classifier from a committee of weak classifiers. The results we achieved using the validation metrics of the New Model 1 showed an overall precision of 69%, a recall of 69%, an F1-Score of 69%, a specificity of 69% and an accuracy of 69% on our diabetes dataset, with an increase of the computation speed by 55 times in comparison of the classical system. Our study has proved that QC improves the computational speed and its inclusion in medical applications will deliver faster results to physicians and caregivers.
3D face reconstruction is an attractive topic in computer vision. We have seen dramatic rise in its development recently. Now the state-of-the-art method can reconstruct a face from a single 2D face image freely, which brings a threat to facial security society. Since they are very similar in feature distributions, an efficient work to discriminate reconstructed face and real face is vital. Since Generative Adversarial Nets (GAN) has been proposed by Ian J. Goodfellow in 2014, it is extensively trained to approximate data distributions of many applications. For its adversarial mechanism, GAN shows a powerful generative ability to get the state of art. Inspired by its adversarial mechanism, we propose a similar framework called Anti-GAN to discriminate an adversarial dataset from real 3D face datasets and reconstructed face datasets. Considering the computation of backpropagation, G and D all adopt convolutional neural network architecture. Additionally, experiments show that Anti-GAN is a powerful way to distinguish real faces and reconstructed faces. At the same time, it can also offer robust features for a facial identity task.
Motivated by the importance of multiple-input and multiple output line-of-sight communication in next generation backhaul networks, in this work we provide an overhead and performance comparison between time- and frequency-domain channel estimation in a bursty 2x2 LOS environment with single-carrier transmission and frequency-domain equalization at the receiver. For both types of channel estimation, analytical expressions for the weights of the involved equalizers are provided in the case of minimum-mean square error equalization. Finally, simulation results are provided regarding their error rate comparison and a discussion concerning their training overhead requirements.
