SVM 기반 음성/음악 분류기의 효율적인 임베디드 시스템 구현

Radio frequency identification (RFID) is widely applied in massive items tagged domains. Existing medium access control (MAC) solutions primarily focus on improving slot efficiency or reducing the total number of slots. However, with pervasive applications of RFID, the time and energy consumption are increasingly important and should be considered in the new design. In this paper, we re-exam the problem of tag identification in UHF RFID system from the perspective of time and energy consumption. The presented work comprehensively reviews and analyzes the prior tag reading protocols. Based on prior art, we further discuss a novel design of tag reading algorithm to improve both time and energy efficiency of EPC C1 Gen2 UHF RFID standard. By exploring the effectiveness of embedding slot-by-slot mechanism in a sub-frame observation phase and combine the sub-frame and slot-by-slot observation in the proposed algorithm, which can achieve more fine-grained frame size adjustment with time and energy-efficiency. Moreover, the cardinality estimation function of the algorithm is implemented by the look-up tables, which allows dramatically reduction in computational complexity and energy consumption. Both simulation results and experiments show clear performance improvement over the commercial solutions.

“Snow is an important component of the terrestrial freshwater budget in high mountain Asia (HMA) and contributes to the runoff in Himalayan rivers through snowmelt. Despite the importance of snow in HMA, considerable spatiotemporal uncertainty exists across the different estimates of snow water equivalent for this region. In order to better estimate snow water equivalent, radiative transfer models are often used in conjunction with microwave brightness temperature measurements. In this study, the efficacy of support vector machines (SVMs), a machine learning technique, to predict passive microwave brightness temperature spectral difference (∆Tb) as a function of geophysical state variables (snow water equivalent, snow depth, snow temperature, and snow density) is explored through a sensitivity analysis. The use of machine learning (as opposed to radiative transfer models) is a relatively new and novel approach for improving snow water equivalent estimates. The Noah-MP land surface model within the NASA Land Information System framework is used to simulate the hydrologic cycle over HMA and model geophysical states that are then used for SVM training. The SVMs serve as a nonlinear map between the geophysical space (modeled in Noah-MP) and the observation space (∆Tb as measured by the radiometer). Advanced Microwave Scanning Radiometer -Earth Observing System measured passive microwave brightness temperatures over snow-covered locations in the HMA region are used as training data during the SVM training phase. Sensitivity of well-trained SVMs to each Noah-MP modeled state variable is assessed by computing normalized sensitivity coefficients. Sensitivity analysis results generally conform with the known first-order physics. Input states that increase volume scattering of microwave radiation, such as snow density and snow water equivalent, exhibit a plurality of positive normalized sensitivity coefficients. In general, snow temperature was the most sensitive input to the SVM predictions. The sensitivity of each state is location and time dependent. The signs of normalized sensitivity coefficients that indicate physical irrationality are ascribed to significant cross-correlation between Noah-MP simulated states and decreased SVM prediction capability at specific locations due to insufficient training data. SVM prediction pitfalls do exist that serve to highlight the limitations of this particular machine learning algorithm.”

Introduction Fatal car accidents have become an alarming issue all over the globe. A sudden medical condition such as a heart attack causes medical symptoms that lead a driver to lose consciousness while driving and consequently leads to a crash. Many studies have demonstrated the high correlation between the driver's sudden medical conditions and involving in a car crash [1][2]. Therefore, to reduce car crashes from the driver's sudden illness from heart-attack as well as save the driver's life in a timely manner, in this work, we discuss the development of a portable wearable system that can continuously monitor the driver for any early symptoms of heart attack and inform him before losing conciuous to stop the car as well as inform medical caregivers to save life. Background Myocardial infarction (MI) is the medical term for the medical condition commonly known as a heart attack, a serious medical emergency in which the blood supply to the heart is suddenly blocked, usually by a blood clot, leading to damage heart muscle [3]. A complete blockage of a coronary artery is a ‘STEMI’ heart attack (ST-elevation MI), whereas a partial blockage would be a ‘NSTEMI’ heart attack (a non-ST-elevationMI) [4]. The average, resting heart rhythm has a QRS-complex following a P-wave and followed by a T-wave, as illustrated in Figure 1(a). A STEMI heart attack will cause an elevation in the ST-complex (Figure 1(b)), whereas a NSTEMI heart attack would not signify ST elevation, but nonetheless can cause ST-segment depression or T-wave inversion (Figure 1(c)), which can be detected immediately by a real-time device to save the driver's life. Method The prototype system consists of two subsystems (Figure 2) that communicate wirelessly using Bluetooth low energy (BLE) technology: wearable sensor subsystem, and an intelligent heart attack detection and warning subsystem. Wearable Subsystem: The wearable chest-belt sub-system includes dry electrodes (reference and two electrodes for differential acquisition), analogue front end (AFE), power management module, and RFDuino microcontroller with BLE. This subsystem acquires the ECG signals from human body continuously and sends these raw measurements wirelessly using BLE technology to the intelligent subsystem. Reusable and smaller dimension dry electrodes (Cognionics, Inc) were embedded in a chest belt to be worn by a car driver. AD82832 AFE is an integrated signal conditioning block to extract, amplify (60 dB gain), and filter (0.48-41 Hz) ECG signal in the presence of noisy conditions. Lithium Polymer (LiPo) battery of 3.7 V (1000 mAH) with the Microchip MCP73831 charge controllers, and Texas instruments' TPS61200 voltage regulators to supply 3 V to the wearable system. The miniaturized ARM Cortex M0 RFDuino microcontroller digitizes the signal at 500 Hz sampling rate and transmits the acquired signal through built-in BLE to decision making subsystem. Intelligent Decision-making Subsystem: This subsystem will receive the ECG signals from the wearable subsystem continuously. It is capable of processing, analyzing the received ECG signals, and making the right decision using support vector machine (SVM) algorithm to classify the normal and abnormal ECG signal to detect heart attack symptoms. This subsystem was built around the single board computer, Raspberry Pi 3 (RPi3) along with SIM 908 GSM and GPS module for location information and alerting service. Multi-threaded python code was written for RPi3 to automatically acquire, buffer, baseline correction and digital smoothing and analyse the ECG data. SVM algorithm was implemented in RPi 3 and used for real-time abnormality detection using the trained model and classification was done using LIBSVM, an open source library [5]. 4-fold cross-validation was used to evaluate classification accuracy. SIM908 GSM+GPS shield attached on the RPi3 to provide car location (latitude, longitude) and to connect to the mobile network for generating an automatic call to medical emergency. This subsystem is designed to take power from the car battery using Cigarette Lighter Socket, which powers the system only when the car's engine is ON. To develop the intelligent program for decision-making subsystem, public MIT-BIH ST change database [6] was used to train a SVM model for normal, ST-elevated, and T-inverted ECG-beats with the time domain (TD), frequency domain (FD) and extended time-frequency domain (TFD) features extracted. The TD features mean, variance, skewness, kurtosis, and coefficient of variation and the FD features spectral flux, spectral entropy and spectral flatness were calculated to spot abnormalities in the ECG-beats. Three time-frequency (TF) distributions were also used in this study: Wigner-Ville Distribution (WVD), Spectrogram (SPEC), and Extended Modified B-Distribution (EMBD). Result and Discussion Recorded ECG Traces: It was clearly revealed from Fig. 5 that the ECG signal transmitted using the prototyped system is in clinical grade. Training SVM: Five hundred traces from each patient and total 2500 traces from MIT-BIH database having either normal or abnormal heart rhythm were segmented and averaged for each case (Figure 6 (A, B, & C)). The power spectral of the signal in Figure 6 (D, E & F) shows that the power spectral density peaks appear at different frequencies for normal and abnormal ECG signals. This reflects that the FD feature can help in classifying the ECG signals. However, TD, FD, and TFD features provide an insight on the signal while compensating for the noise or motion artefacts. Classification using SVM: Table 1 below summarizes the accuracy of the prototyped device. EMBD produces higher accuracy in classification of ECG signal. Conclusion This work shows the possibility to detect driver's heart attack reliably using the developed prototype system. SVM machine learning algorithm that was trained with a sufficiently high number of training data can classify STEMI or NSTEMI with approximately 97.4% and 96.3% accuracy respectively when the extended TF features (with EMBD distribution) were used for training and classification. The maximum current drawn by the wearable chest-belt subsystem during continuous acquisition is 9.3 mA, which ensures the life span of a 1000 mAh LiPo battery is 75 hours, once it is fully charged and therefore it can be expected that the device can run longer without requiring recharging daily.

A multi-layer soil moisture data assimilation using support vector machines and ensemble particle filter

Support vector machines (SVM) have been shown to outperform classical supervised classification algorithms, and have therefore been recently used for classification of hyperspectral images. This paper present hyperspetral image classification based on support vector machines with two different unsupervised pre-segmentation methods applied to hyperspectral training data before the training phase of SVM classification. The pre-segmentation step, in a way compresses the training data by combining similar hyperspectral data, as a result of which the number of training samples is reduced, resulting in an overall smaller support vector amount after training. In this paper, compression is achieved using kmeans and phase correlation based unsupervised segmentation methods before the SVM training phase. It is shown that with the proposed approach it is possible to trade of accuracy against sparsity and also provide faster training time. Sparsity is important, particularly considering the high data amount encountered in hyperspectral imaging, because sparsity determines the model complexity and therefore the computational burden of the classification phase.

The current development of Support Vector Machines (SVM) has reached a bottleneck, with issues such as long training times and weak interpretability when dealing with large-scale, multi-dimensional data. This paper introduces the concept of Quantum Support Vector Machines (QSVM) and achieves efficient solutions through quantum algorithms such as the HHL algorithm. The research designs a computational architecture that combines classical and quantum computing, utilizing the Pauli decomposition of Hermitian matrices to simulate the quantum simulation of Hamiltonian quantities and implementing the quantum simulation of unitary operators. Based on the conclusions, a complete quantum linear solver circuit is designed, achieving exponential growth in computational complexity. Experiments using the Iris dataset demonstrate the excellent classification performance of QSVM, which outperform classical SVM in classification accuracy, computational time complexity, and memory requirements. More efficient quantum mapping algorithms and quantum circuit optimization methods are implemented, providing new ideas and methods for the application of SVM to large-scale datasets.

近年来，支持向量机(SVM)理论广泛应用于模式分类，然而影响其分类准确率的两个主要因素特征选择和参数优化又是相互影响和制约的。文章提出一种BA + SVM算法，利用蝙蝠算法(BA)来同步完成SVM的参数优化和输入数据的特征属性选择，提高了SVM的分类能力。设计的三种实验方式在10个测试数据集上实验结果表明，BA + SVM同步优化算法与单一进行参数优化或单一进行特征选择算法相比，具有输入特征少分类准确率高的优势。 Support vector machines (SVM), which is a popular method for pattern classification, has been recently adopted in range of problems. In training procedure of SVM, feature selection and parameter optimization are two main factors that impact on classification accuracy. In order to improve the classification accuracy by optimizing parameter and choosing feature subset for SVM, a new algorithm is proposed through combining Bat Algorithm (BA) with SVM, termed BA + SVM. For assessing the performance of BA + SVM, 10 public data-sets are employed to test the classification accuracy rate. Compared with grid algorithm, conventional parameter optimization method, our study concludes that BA + SVM has a higher classification accuracy with fewer input features for support vector classification.

Recent advancements in maritime technologies, pure oceanic research and military techniques related to underwater acoustics have introduced us to a new class of problems in digital signal processing. Apart from acoustic measurements, underwater classification to identify signatures poses a challenging task at hand due to typical profile of anthropogenic noise which is very complex. Moreover, unavailability of large relevant datasets makes it hard to achieve high accuracy in classification. Underwater acoustic channel is a dynamic one that not only includes noise due to fish and marine life but also that from commercial human activity and military operations. This work shows that a single classification approach is not enough to classify signals with very high accuracy. Several techniques of underwater acoustic classification using objects' unique acoustic signature have been investigated. This classification process using acoustic signatures is divided into two stages. In the first stage, preprocessing and feature extraction needs to be performed on the given dataset — which typically includes a variety of marine life and anthropogenic signals. In the classification stage, multiple techniques including Naive Bayesian, K nearest neighbors, Artificial neural networks, Support vector machines and Hidden Markov Models have been performed to evaluate classification results. The shortcomings of each classifier are discussed thoroughly in this paper. The main challenges for the robust and fast classification of underwater signals are also described to mitigate them in future. Moreover, open research issues are also discussed, and possible solution approaches are outlined.

How do temporal and spectral features matter in crop classification in Heilongjiang Province, China?

Because of the plague that threatens the entire planet, people continue to live in dread and instability. COVID-19 is considered one of the fastest pandemics that affected the whole world. As an initial step, PCR analysis is becoming more and more popular, although it has issues with accuracy. Using Convolutional neural network (CNN), it takes a while to identify a few computers tomography (CT) images despite its great accuracy in proper classification Additionally, classification using Data-Mining algorithms on (CT) images has been researched, and they achieve high accuracy in classification, but it is time-consuming. While scanning (CT) images, Patients are in danger when utilizing imaging equipment. Even after a deep sterilization, there is a major likelihood that the infection will persist on the scanning chamber surface.This research presents a COVID-19 patient diagnosis approach apply to find a quickly and so effective classification using a blood test. The suggested approach consists of two main phases: a feature selection phase and covid-19 prediction phase. The feature selection is applied by implementing The Chi-Square feature selection technique. This methodology, a filter feature selection technique, can swiftly identify the most useful subset of features. Then, an improved support vector machine is used to deliver an immediate and accurate diagnosis (RBF SVM). The basic concept behind the suggested RBF SVM is choosing the value of hyperparameters, C and Y, in the supporting vector machine (SVM) formula before computing the SVM. The results demonstrate that the suggested method provides an accuracy of up to 98.8%.

“Support vector machine (SVM)” is one of the well-known machine learning algorithms used for classification and regression tasks. The “SVM” parameters and features selection have a significant impact on the “SVM” model's complexity and classification accuracy. Finding the best set of features and selecting the appropriate parameters of “SVM” is considered an optimization problem. Different metaheuristic algorithms are employed in the literature to choose the best set of features and to optimize “SVM” parameters simultaneously. This paper presents a new algorithm hybrid based on “Aquila Optimizer” and “Whale Optimization Algorithm” with Adaptive inertial weight, called AOWOA. The AOWOA is used for global optimization and feature selection and optimizing “SVM” parameters to achieve a higher classification accuracy. We utilized two experiment series to test the presented (AOWOA) algorithm. In the first experiment, 13 standard “benchmark functions” are used and the AOWOA algorithm is compared to AO, MVO, WOA, MFO, PSO, and SSA. In the second experiment, the results of AOWOA-SVM are compared with four metaheuristic algorithms: MVO, GWO, WOA, and BAT using ten labeled datasets. The experiment results proved that AOWOA can reduce the number of features, find the optimal parameters of “SVM”, and avoids local optima with high classification accuracy in most datasets compared to other algorithms.

Website accumulates a large number of customer reviews for goods and website services. Support vector machine (SVM) is an effective text categorization method, it has strong generalization ability and high classification accuracy which can be used to track and manage customer reviews. But SVM has some weaknesses which slow training convergence speed and difficult to raise the classification accuracy. The paper use heterogeneous kernel functions which have different characteristics to resolve the problem of SVM weak generalization ability to learn and improve the SVM classification accuracy. Through classify customer reviews, online shopping websites resolve issues of critical analysis about mass customers reviews and effectively improve website service standard. Key words : Customer Review; Text Categorization; SVM; Multiple Kernel Learning

Clustering technique-based least square support vector machine for EEG signal classification

Performance and energy impact of OpenMP runtime configurations on power constrained systems

Even though support vector machine has been successfully applied to various classification problems with its flexibility and high classification accuracy, it is not suitable for classification of large data sets because its computational complexity grows rapidly as the size of data set increases. SVM using k-means clustering (KM-SVM) is a fast algorithm which has been developed to accelerate both computation and prediction of SVM classifiers. However, it seems likely that the data set is contaminated by outliers in real-world situations, and k-means clustering is sensitive to these outliers. Therefore, we propose to combine k-spatial medians clustering with SVM (KS-SVM) since k-spatial medians clustering is robust for outliers. In order to improve the classification accuracy in KS-SVM, furthermore, a recovery process based on KS-SVM (RKS-SVM) is also proposed in this article. Experiments show that KS-SVM can improve the performance of KM-SVM in terms of classification accuracy and number of support vector. It is also shown that the classification accuracy of RKS-SVM is better than one of KS-SVM, but there is a tradeoff between classification accuracy and the number of support vectors.