Search

Compared with other applications in computer vision, convolutional neural networks (CNNs) have underperformed on pedestrian detection. A breakthrough was made very recently using sophisticated deep CNN (DCNN) models, with a number of handcrafted features or explicit occlusion handling mechanism. In this paper, we show that by reusing the convolutional feature maps of a DCNN model as image features to train an ensemble of boosted decision models, we are able to achieve the best reported accuracy without using specially designed learning algorithms. We empirically identify and disclose important implementation details. We also show that pixel labeling may be simply combined with a detector to boost the detection performance. By adding complementary handcrafted features such as optical flow, the DCNN-based detector can be further improved. We advance the state-of-the-art results by lowering the log-average miss rate from 11.7% to 8.9% on the Caltech data set and from 11.2% to 8.6% on the Inria data set. We also achieve a comparable result to state-of-the-art approaches on the KITTI data set. [ABSTRACT FROM AUTHOR]

Under-complete models, which derive lower dimensional representations of input data, are valuable in domains in which the number of input dimensions is very large, such as data consisting of a temporal sequence of images. This paper presents the under-complete product of experts (UPoE), where each expert models a one-dimensional projection of the data. Maximum-likelihood learning rules for this model constitute a tractable and exact algorithm for learning under-complete independent components. The learning rules for this model coincide with approximate learning rules proposed earlier for under-complete independent component analysis (UICA) models. This paper also derives an efficient sequential learning algorithm from this model and discusses its relationship to sequential independent component analysis (ICA), projection pursuit density estimation, and feature induction algorithms for additive random field models. This paper demonstrates the efficacy of these novel algorithms on high-dimensional continuous datasets. [ABSTRACT FROM AUTHOR]

In this paper, we focus on tackling the problem of automatic accurate localization of detected objects in high-resolution remote sensing images. The two major problems for object localization in remote sensing images caused by the complex context information such images contain are achieving generalizability of the features used to describe objects and achieving accurate object locations. To address these challenges, we propose a new object localization framework, which can be divided into three processes: region proposal, classification, and accurate object localization process. First, a region proposal method is used to generate candidate regions with the aim of detecting all objects of interest within these images. Then, generic image features from a local image corresponding to each region proposal are extracted by a combination model of 2-D reduction convolutional neural networks (CNNs). Finally, to improve the location accuracy, we propose an unsupervised score-based bounding box regression (USB-BBR) algorithm, combined with a nonmaximum suppression algorithm to optimize the bounding boxes of regions that detected as objects. Experiments show that the dimension-reduction model performs better than the retrained and fine-tuned models and the detection precision of the combined CNN model is much higher than that of any single model. Also our proposed USB-BBR algorithm can more accurately locate objects within an image. Compared with traditional features extraction methods, such as elliptic Fourier transform-based histogram of oriented gradients and local binary pattern histogram Fourier, our proposed localization framework shows robustness when dealing with different complex backgrounds. [ABSTRACT FROM PUBLISHER]

This paper presents a novel intelligent autoreclosure technique to discriminate temporary faults from permanent faults, and to accurately determine fault extinction time. A variety of fault simulations is carried out on a specified transmission line on the standard IEEE 9-bus electric power system by using MATLAB/SimPowerSytems. Prony analysis is employed to extract data features from each simulated fault. The fault identification prior to reclosing is accomplished by an artificial neural network trained by Levenberg Marquardt and resilient backpropagation algorithms, which are developed by using MATLAB. Some important parameters which strongly affect the entire training process are fine-tuned to their corresponding best values with the help of Taguchi's method. Test results show the robustness and efficacy of the proposed autoreclosure scheme. [ABSTRACT FROM AUTHOR]

A novel framework using a Bayesian approach for content-based phishing web page detection is presented. Our model takes into account textual and visual contents to measure the similarity between the protected web page and suspicious web pages. A text classifier, an image classifier, and an algorithm fusing the results from classifiers are introduced. An outstanding feature of this paper is the exploration of a Bayesian model to estimate the matching threshold. This is required in the classifier for determining the class of the web page and identifying whether the web page is phishing or not. In the text classifier, the naive Bayes rule is used to calculate the probability that a web page is phishing. In the image classifier, the earth mover's distance is employed to measure the visual similarity, and our Bayesian model is designed to determine the threshold. In the data fusion algorithm, the Bayes theory is used to synthesize the classification results from textual and visual content. The effectiveness of our proposed approach was examined in a large-scale dataset collected from real phishing cases. Experimental results demonstrated that the text classifier and the image classifier we designed deliver promising results, the fusion algorithm outperforms either of the individual classifiers, and our model can be adapted to different phishing cases. [ABSTRACT FROM AUTHOR]

In this paper, we present a novel feature-based neural network (NN) approach for estimation of blood pressure (BP) from wrist oscillometric measurements. Unlike previous methods that use the raw oscillometric waveform envelope (OMWE) as input to the NN, in this paper, we propose to use features extracted from the envelope. The OMWE is mathematically modeled as a sum of two Gaussian functions. The optimum parameters of this model are found by minimizing the least squares error between the model and the OMWE using the Levenberg–Marquardt algorithm and are used as features. Two separate feed-forward NNs (FFNNs) are then designed to estimate the systolic and diastolic BPs using these features. The FFNNs are trained using the resilient backpropagation learning algorithm and tested on a data set of BP measurements recorded from 85 subjects. The performance is then compared with that of the conventional maximum amplitude algorithm, adaptive neuro-fuzzy inference system, and already published NN-based methods. It is found that the proposed approach achieves lower values of mean absolute error and standard deviation of error in the estimation of BP. In addition, the proposed approach has the following advantages: lower complexity with respect to the design parameters, smaller training data set, and lower computational load. [ABSTRACT FROM PUBLISHER]

Virtual network embedding arranges virtual network services onto substrate network components. The performance of embedding algorithms determines the effectiveness and efficiency of a virtualized network, making it a critical part of the network virtualization technology. To achieve better performance, the algorithm needs to automatically detect the network status which is complicated and changes in a time-varying manner, and to dynamically provide solutions that can best fit the current network status. However, most existing algorithms fail to provide automatic embedding solutions in an acceptable running time. In this paper, we combine deep reinforcement learning with a novel neural network structure based on graph convolutional networks, and propose a new and efficient algorithm for automatic virtual network embedding. In addition, a parallel reinforcement learning framework is used in training along with a newly-designed multi-objective reward function, which has proven beneficial to the proposed algorithm for automatic embedding of virtual networks. Extensive simulation results under different scenarios show that our algorithm achieves best performance on most metrics compared with the existing state-of-the-art solutions, with upto 39.6% and 70.6% improvement on acceptance ratio and average revenue, respectively. Moreover, the results also demonstrate that the proposed solution possesses good robustness. [ABSTRACT FROM AUTHOR]

The shortage of training samples remains one of the main obstacles in applying the neural networks to the hyperspectral images classification. To fuse the spatial and spectral information, pixel patches are often utilized to train a model, which may further aggregate this problem. In the existing works, an artificial neural network (ANN) model supervised by centerloss (ANNC) was introduced. Training merely with spectral information, the ANNC yields discriminative spectral features suitable for the subsequent classification tasks. In this paper, we propose a novel convolutional neural network (CNN)-based spatial feature fusion (CSFF) algorithm, which allows a smart integration of spatial information to the spectral features extracted by ANNC. As a critical part of CSFF, a CNN-based discriminant model is introduced to estimate whether two pixels belong to the same class. At the testing stage, by applying the discriminant model to the pixel pairs generated by a test pixel and each of its neighbors, the local structure is estimated and represented as a customized convolutional kernel. The spectral–spatial feature is generated by a convolutional operation between the estimated kernel and the corresponding spectral features within a local region. The final label is determined by classifying the resulting spectral–spatial feature. Without increasing the number of training samples or involving pixel patches at the training stage, the CSFF framework achieves the state of the art by declining 20%–50% classification failures in experiments on three well-known hyperspectral images. [ABSTRACT FROM AUTHOR]

A fast target maneuver detecting and highly accurate tracking technique using a neural fuzzy network based on Kalman filter is proposed in this paper. In the automatic target tracking system, there exists an important and difficult problem: how to detect the target maneuvers and fast response to avoid miss-tracking? The traditional maneuver detection algorithms, such as variable dimension filter (VDF) and input estimation (IE) etc., are computation intensive and difficult to implement in real time. To solve this problem, neural network algorithms have been issued recently. However, the normal neural networks such as backpropagation networks usually produce the extra problems of low convergence speed and/or large network size. Furthermore, the way to decide the network structure is heuristic. To overcome these defects and to make use of neural learning ability, a developed standard Kalman filter with a self-constructing neural fuzzy inference network (KF-SONFIN) algorithm for target tracking is presented in this paper. By generating possible target trajectories including maneuver information to train the SONFIN, the trained SONFIN can detect when the maneuver occurred, the magnitude of maneuver values and when the maneuver disappeared. Without having to change the structure of Kalman filter nor modeling the maneuvering target, this new algorithm, SONFIN, can always find itself an economic network size with a fast learning process. Simulation results show that the KF-SONFIN is superior to the traditional IE and VDF methods in estimation accuracy. [ABSTRACT FROM AUTHOR]

Partial discharge (PD) recognition is an important tool for online monitoring of gas insulated switchgear (GIS) and diagnosing existing defects. At present, there are two different types of data patterns used for analysis and evaluation of PD signals: phase resolved partial discharge (PRPD) mode and time resolved partial discharge (TRPD) mode. Using different types of data patterns separately can lead to inconsistent or even conflicted recognition results, but the two types of data patterns having complementary information between each other can be used together for data fusion. Dempster-Shafer (DS) evidence theory is introduced to address the problem of evidence conflict and low fusion efficiency. First, two sub-networks for PD recognition are established and compared employing the back propagation neural network (BPNN) learning algorithm under the two modes respectively. Secondly, to minimize (with possible elimination of) the possibility of misclassification, a new fusion decision-making system for PD recognition is proposed on the basis of fusing the results from the sub-networks. Finally, extensive field experiments on two artificial defects are conducted in order to evaluate the performance of the proposed method in this paper. The classification results reveal that the proposed method significantly outperforms the ways of using only single type of data patterns. The method proposed in this paper reduces the uncertainty and effectively improves the credibility of diagnosis. [ABSTRACT FROM AUTHOR]

Object detection plays an important role in intelligent transportation systems and intelligent vehicles. Although the topic of object detection has been studied for decades, it is still challenging to accurately detect objects under complex scenarios. The contributing factors for challenges include diversified object and background appearance, motion blur, adverse weather conditions, and complex interactions among objects. In this paper, we propose a new convolutional neural network (CNN) model for traffic object detection, by using multi-scale local and global feature representation (MFR). The proposed model consists of two components: a region proposal network that generates candidate object regions and an object detection network that incorporates multi-scale features and global information, namely MFR-CNN. These two components are jointly optimized. Once the system is trained, it can detect real-world traffic objects accurately, especially small objects and heavily occluded objects. We evaluate the proposed method on four benchmark datasets, achieving consistent improvements over the state of the art. [ABSTRACT FROM AUTHOR]

This paper presents a neural network-based method to generate multiple images with different exposures from a single input low dynamic range (LDR) image for improved high dynamic range (HDR) imaging. The proposed algorithm consists of three steps: 1) 2-D histogram estimation; 2) neural network-based LDR images estimation; and 3) generation of an optimal set of differently exposed images. The proposed method first generates image features by estimating a patched-based 2-D histogram. The extracted features are used in an input layer of the neural network, which plays a role to select an optimal set of LDR images. A set of LDR images is generated using a curvature-based contrast enhancement method. Experimental results show that the proposed method can generate an optimal set of LDR images using neural network and provide improved HDR images. In addition, the proposed method can be implemented as a preprocessing step in most existing HDR frameworks. The proposed HDR approach is considered as a single-input method that gives almost the same performance to multiple image-based HDR method. [ABSTRACT FROM PUBLISHER]

Following the great success of deep convolutional neural networks (CNNs) in computer vision, this paper proposes a complex-valued CNN (CV-CNN) specifically for synthetic aperture radar (SAR) image interpretation. It utilizes both amplitude and phase information of complex SAR imagery. All elements of CNN including input-output layer, convolution layer, activation function, and pooling layer are extended to the complex domain. Moreover, a complex backpropagation algorithm based on stochastic gradient descent is derived for CV-CNN training. The proposed CV-CNN is then tested on the typical polarimetric SAR image classification task which classifies each pixel into known terrain types via supervised training. Experiments with the benchmark data sets of Flevoland and Oberpfaffenhofen show that the classification error can be further reduced if employing CV-CNN instead of conventional real-valued CNN with the same degrees of freedom. The performance of CV-CNN is comparable to that of existing state-of-the-art methods in terms of overall classification accuracy. [ABSTRACT FROM PUBLISHER]

In recent years, many alternative methodologies and techniques have been proposed to perform nondestructive inspection and maintenance operations of moving structures. In particular, ultrasonic techniques have shown to be very promising for automatic inspection systems. From the literature, it is evident that the neural paradigms are considered, by now, the best choice to automatically classify ultrasound data. At the same time, the most appropriate preprocessing technique is still undecided. The aim of this paper is to propose a new and innovative data preprocessing technique that converts real-valued ultrasonic data into complex-valued signals. This allows analysis using phase synchrony, a robust tool that has been previously employed in brain science for establishing robust features in noisy data. Synchrony estimation is achieved using complex extensions of empirical mode decomposition, a data-driven algorithm for detecting temporal scales, thus facilitating the modeling of nonlinear and nonstationary signal dynamics. Experimental tests aiming to detect defective areas in composite materials are reported, and the effectiveness of the proposed methodology is illustrated. [ABSTRACT FROM PUBLISHER]

This paper presents a novel intelligent autoreclosure technique to discriminate temporary faults from permanent faults and accurately determine fault extinction time. A variety of fault simulations are carried out on a specified transmission line on the standard IEEE 9-bus electric power system using MATLAB/SimPowerSytems. FFT and Prony analysis methods are employed to extract data features from each simulated fault. The fault identification prior to reclosing is accomplished by an artificial neural network trained by standard Error Back-Propagation, Levenberg Marquardt, and Resilient Back-Propagation algorithms which are developed using MATLAB. Some important parameters which strongly affect the entire training process are fine tuned with Taguchi's method to their corresponding best values. The robustness of the developed ANN identifier is verified by testing it with the data patterns which consists of high impedance faults obtained from IEEE 14-bus benchmark system. Test results show the efficacy of the proposed AR scheme. [ABSTRACT FROM AUTHOR]

This paper presents the Visual Attention Engine (VAE), which is a digital cellular neural network (CNN) that executes the VA algorithm to speed up object-recognition. The proposed time-multiplexed processing element (TMPE) CNN topology achieves high performance and small area by integrating 4800 (80\,\times\,60) cells and 120 PEs. Pipelined operation of the PEs and single-cycle global shift capability of the cells result in a high PE utilization ratio of 93%. The cells are implemented by 6T static random access memory-based register files and dynamic shift registers to enable a small area of 4.5 mm^2. The bus connections between PEs and cells are optimized to minimize power consumption. The VAE is integrated within an object-recognition system-on-chip (SoC) fabricated in the 0.13-\mum complementary metal-oxide—semiconductor process. It achieves 24 GOPS peak performance and 22 GOPS sustained performance at 200 MHz enabling one CNN iteration on an 80\,\times\,60 pixel image to be completed in just 4.3 \mus. With VA enabled using the VAE, the workload of the object-recognition SoC is significantly reduced, resulting in 83% higher frame rate while consuming 45% less energy per frame without degradation of recognition accuracy. [ABSTRACT FROM AUTHOR]

In this paper, we propose an efficient general-purpose no-reference (NR) video quality assessment (VQA) framework that is based on 3D shearlet transform and convolutional neural network (CNN). Taking video blocks as input, simple and efficient primary spatiotemporal features are extracted by 3D shearlet transform, which are capable of capturing natural scene statistics properties. Then, CNN and logistic regression are concatenated to exaggerate the discriminative parts of the primary features and predict a perceptual quality score. The resulting algorithm, which we name shearlet- and CNN-based NR VQA (SACONVA), is tested on well-known VQA databases of Laboratory for Image & Video Engineering, Image & Video Processing Laboratory, and CSIQ. The testing results have demonstrated that SACONVA performs well in predicting video quality and is competitive with current state-of-the-art full-reference VQA methods and general-purpose NR-VQA algorithms. Besides, SACONVA is extended to classify different video distortion types in these three databases and achieves excellent classification accuracy. In addition, we also demonstrate that SACONVA can be directly applied in real applications such as blind video denoising. [ABSTRACT FROM PUBLISHER]

In this study, seventeen samples were created for classifying internal, surface, and corona partial discharges (PDs) in a high voltage lab. Next, PDs were measured experimentally to provide a dictionary comprising the types. Due to the huge size of the recorded dataset, a new and straightforward preprocessing method based on signal norms was used to extract the appropriate features of various samples. The new sparse representation classifier (SRC) was computed using ?1 and stable ?1-norm minimization by means of Primal-Dual Interior Point (PDIP) and Basis Pursuit De-noise (BPDN) algorithms, respectively. The pattern recognition was also performed with an artificial neural network (ANN) and compared with the sparse method. It is shown that both methods have comparable performance if training process, tuning options, and other tasks for finding the best result from ANN are not taken into account. Even with this assumption, it is shown that SRC still performs better than ANN in some cases. In addition, the SRC technique presented in this paper converges to a fixed result, while the results after training the ANN vary with every run due to random initial weights. [ABSTRACT FROM AUTHOR]

A convolutional neural network (CNN) has recently demonstrated its outstanding capability for the classification of hyperspectral images (HSIs). Typical CNN-based methods usually adopt image patches as inputs to the network. However, a fixed-size image patch in HSI with complex spatial contexts may contain multiple ground objects of different classes, which will deteriorate the classification performance of the CNN. In addition, traditional convolutional layers adopted in the CNN have a huge amount of parameters needed to be tuned, which will cause high computational cost. To address the above-mentioned issues, a novel squeeze multibias network (SMBN) is proposed for HSI classification. Specifically, the proposed SMBN first introduces the multibias module (MBM), which incorporates multibias into the rectified linear unit layers. The MBM can decouple the feature maps of input patches into multiple response maps (corresponding to different ground objects) and adaptively select the meaningful maps for classification. Furthermore, the proposed SMBN replaces the traditional convolutional layer with a squeeze convolution module, which can greatly reduce the number of parameters in the network, thus saving the running time, while still maintaining high classification accuracy. Experimental results on three real HSIs demonstrate the superiority of the proposed SMBN method over several state-of-the-art classification approaches. [ABSTRACT FROM AUTHOR]

This paper proposes an optimal feature selection approach, namely, probabilistic neural network-based feature selection (PFS), for power-quality disturbances classification. The PFS combines a global optimization algorithm with an adaptive probabilistic neural network (APNN) to gradually remove redundant and irrelevant features in noisy environments. To validate the practicability of the features selected by the proposed PFS approach, we employed three common classifiers: multilayer perceptron, k-nearest neighbor and APNN. The results indicate that this PFS approach is capable of efficiently eliminating nonessential features to improve the performance of classifiers, even in environments with noise interference. [ABSTRACT FROM AUTHOR]

Glaucoma is the second leading cause of blindness worldwide. It is a disease in which fluid pressure in the eye increases continuously, damaging the optic nerve and causing vision loss. Computational decision support systems for the early detection of glaucoma can help prevent this complication. The retinal optic nerve fiber layer can be assessed using optical coherence tomography, scanning laser polarimetry, and Heidelberg retina tomography scanning methods. In this paper, we present a novel method for glaucoma detection using a combination of texture and higher order spectra (HOS) features from digital fundus images. Support vector machine, sequential minimal optimization, naive Bayesian, and random-forest classifiers are used to perform supervised classification. Our results demonstrate that the texture and HOS features after z-score normalization and feature selection, and when combined with a random-forest classifier, performs better than the other classifiers and correctly identifies the glaucoma images with an accuracy of more than 91%. The impact of feature ranking and normalization is also studied to improve results. Our proposed novel features are clinically significant and can be used to detect glaucoma accurately. [ABSTRACT FROM AUTHOR]

Two of the most critical tasks when designing a portable wireless neural recording system are to limit power consumption and to efficiently use the limited bandwidth. It is known that for most wireless devices the majority of power is consumed by the wireless transmitter and it often represents the bottleneck of the overall design. This paper compares two compression techniques that take advantage of the sparseness of the neural spikes in neural recordings using an information theoretic formalism to enhance the well-established vector quantization (VQ) algorithm. The two discriminative VQ algorithms are applied to neuronal recordings proving their ability to accurately reconstruct action potential (AP) regions of the neuronal signal while compressing background activity without using thresholds. The two operational modes presented offer distinct characteristics to lossy compression. The first approach requires no preprocessing or prior knowledge of the signal while the second requires a training set of spikes to obtain AP templates. The compression algorithms are implemented on an on-board digital signal processor (DSP) and results show that power consumption is decreased while the bandwidth is more efficiently utilized. The compression algorithms have been tested in real time on a hardware platform (PICO DSP refid="ref11"/) enhanced with the DSP which runs the algorithm before sending the compressed data to a wireless transmitter. The compression ratios obtained range from 70:1 and 40:1 depending on the signal to noise ratio (SNR) of the input signal. The spike sorting accuracy in the reconstructed data is 95% compatible to the original neural data. [ABSTRACT FROM PUBLISHER]

We propose in this letter two new subspace learning methods, called nearest feature space analysis (NFSA) and discriminant nearest feature space analysis (DNFSA), for pattern classification. While many subspace learning algorithms have been proposed in recent years, most of them apply the conventional nearest neighbor (NN) metric to derive the subspace and may not effectively characterize the geometrical information of the samples, especially when the number of training samples per class is limited. In this paper, we propose using the nearest feature space (NFS) metric to seek a NFSA subspace to improve the discriminating power of the subspace for classification. To further enhance the discriminative power of NFSA, we also propose a new DNFSA method to minimize the within-class feature space (FS) distances and maximize the between-class FS distances simultaneously in the derived subspace. Experimental results on face and facial expression recognition are presented to demonstrate the efficacy of the proposed methods. [ABSTRACT FROM PUBLISHER]

Dimensionality reduction and incremental learning have recently received broad attention in many applications of data mining, pattern recognition, and information retrieval. Inspired by the concept of manifold learning, many discriminant embedding techniques have been introduced to seek low-dimensional discriminative manifold structure in the high-dimensional space for feature reduction and classification. However, such graph-embedding framework-based subspace methods usually confront two limitations: 1) since there is no available updating rule for local discriminant analysis with the additive data, it is difficult to design incremental learning algorithm and 2) the small sample size (SSS) problem usually occurs if the original data exist in very high-dimensional space. To overcome these problems, this paper devises a supervised learning method, called local discriminant subspace embedding (LDSE), to extract discriminative features. Then, the incremental-mode algorithm, incremental LDSE (ILDSE), is proposed to learn the local discriminant subspace with the newly inserted data, which applies incremental learning extension to the batch LDSE algorithm by employing the idea of singular value-decomposition (SVD) updating algorithm. Furthermore, the SSS problem is avoided in our method for the high-dimensional data and the benchmark incremental learning experiments on face recognition show that ILDSE bears much less computational cost compared with the batch algorithm. [ABSTRACT FROM PUBLISHER]

In the paper, the authors propose a security monitoring system that can detect and classify the location and nature of different sounds within a room. This system is reliable and robust even in the presence of reverberation and in low signal-to-noise (SNR) environments. We describe a novel algorithm for audio classification, which, first, classifies an audio segment as speech or nonspeech and, second, classifies nonspeech audio segments into a particular audio type. To classify an audio segment as speech or nonspeech, this algorithm divides the audio segment into frames, estimates the presence of pitch in each frame, and calculates a pitch ratio (PR) parameter; it is this PR parameter that is used to discriminate speech audio segments from nonspeech audio segments. The discerning threshold for the PR parameter is adaptive to accommodate different environments. A time-delayed neural network is employed to further classify nonspeech audio segments into an audio type. The performance of this novel audio classification algorithm is evaluated using a library of audio segments. This library includes both speech segments and nonspeech segments, such as windows breaking and footsteps. Evaluation is performed under different SNR environments, both with and without reverberation. Using 0.4-s audio segments, the proposed algorithm can achieve an average classification accuracy of 94.5% for the reverberant library and 95.1% for the nonreverberant library. [ABSTRACT FROM AUTHOR]

An ever-increasing number of computer vision and image/video processing challenges are being approached using deep convolutional neural networks, obtaining state-of-the-art results in object recognition and detection, semantic segmentation, action recognition, optical flow, and super resolution. Hardware acceleration of these algorithms is essential to adopt these improvements in embedded and mobile computer vision systems. We present a new architecture, design, and implementation, as well as the first reported silicon measurements of such an accelerator, outperforming previous work in terms of power, area, and I/O efficiency. The manufactured device provides up to 196 GOp/s on 3.09 \text mm^2 of silicon in UMC 65-nm technology and can achieve a power efficiency of 803 GOp/s/W. The massively reduced bandwidth requirements make it the first architecture scalable to TOp/s performance. [ABSTRACT FROM AUTHOR]

This paper proposes a novel method to detect anomalies in network traffic, based on a nonrestricted \alpha-stable first-order model and statistical hypothesis testing. To this end, we give statistical evidence that the marginal distribution of real traffic is adequately modeled with \alpha-stable functions and classify traffic patterns by means of a Generalized Likelihood Ratio Test (GLRT). The method automatically chooses traffic windows used as a reference, which the traffic window under test is compared with, with no expert intervention needed to that end. We focus on detecting two anomaly types, namely floods and flash-crowds, which have been frequently studied in the literature. Performance of our detection method has been measured through Receiver Operating Characteristic (ROC) curves and results indicate that our method outperforms the closely-related state-of-the-art contribution described in [CHECK END OF SENTENCE]. All experiments use traffic data collected from two routers at our university—a 25,000 students institution—which provide two different levels of traffic aggregation for our tests (traffic at a particular school and the whole university). In addition, the traffic model is tested with publicly available traffic traces. Due to the complexity of \alpha-stable distributions, care has been taken in designing appropriate numerical algorithms to deal with the model. [ABSTRACT FROM AUTHOR]

Electricity theft behaviors have caused great harm to the economic benefits of power companies and the secure operation of power systems, thus electricity theft detection is paid much attention in the actual power supply management. In this work, a two-step electricity theft detection strategy is proposed to identify electricity theft users and predict potentially stolen electricity (PSE) for maximizing economic return. In the first step, a neural network model called convolutional autoencoder (CAE) is proposed for electricity theft identification, and the convolutional layer is adopted in CAE to extract and identify the abnormalities of electricity theft users against the uniformity and periodicity of normal power consumption features. In the second step, the PSE of each identified electricity theft user is predicted by the improved regression algorithm named Tr-XGBoost, which combines the extreme gradient boosting (XGBoost) algorithm and transfer adaptive boosting (TrAdaBoost) training strategy. The propsoed Tr-XGBoost could learn the relationship between the extracted electricity features and the PSE of each electricity theft user, and then the predicted PSE can be used to determine the list of electricity theft users to be inspected for maximizing economic return. Case studies on both the IEEE 33-bus test system and a low-voltage distribution system of a province in China show that the proposed two-step electricity theft detection strategy can improve the accuracy of electricity theft identification, and obtain a larger economic return because of a more accurate result of PSE prediction than other state-of-the-art algorithms. [ABSTRACT FROM AUTHOR]

Parkinson's Disease (PD) is a slowly evolving neurological disease that affects about $\text{1}\%$ of the population above 60 years old, causing symptoms that are subtle at first, but whose intensity increases as the disease progresses. Automated detection of these symptoms could offer clues as to the early onset of the disease, thus improving the expected clinical outcomes of the patients via appropriately targeted interventions. This potential has led many researchers to develop methods that use widely available sensors to measure and quantify the presence of PD symptoms such as tremor, rigidity and braykinesia. However, most of these approaches operate under controlled settings, such as in lab or at home, thus limiting their applicability under free-living conditions. In this work, we present a method for automatically identifying tremorous episodes related to PD, based on IMU signals captured via a smartphone device. We propose a Multiple-Instance Learning approach, wherein a subject is represented as an unordered bag of accelerometer signal segments and a single, expert-provided, tremor annotation. Our method combines deep feature learning with a learnable pooling stage that is able to identify key instances within the subject bag, while still being trainable end-to-end. We validate our algorithm on a newly introduced dataset of 45 subjects, containing accelerometer signals collected entirely in-the-wild. The good classification performance obtained in the conducted experiments suggests that the proposed method can efficiently navigate the noisy environment of in-the-wild recordings. [ABSTRACT FROM AUTHOR]

Recently, deep learning techniques have been introduced to address hyperspectral image (HSI) classification problems and have achieved the state-of-the-art performances. In this article, we propose a novel clustering algorithm for HSI based on learning embedding using the set-to-set and sample-to-sample distances (LSSDs). This technique consists of four main components: 1) oversegmentation; 2) generation of set-to-set and sample-to-sample distances; 3) learning embedding by training a siamese network; and 4) density-based spectral clustering. First, the HSI is oversegmented into superpixels by using the entropy rate superpixel (ERS) algorithm. Second, the set-to-set distances are obtained by representing the segmented sets of samples as affine hull (AH) models, whereas the sample-to-sample distances are computed by employing the local covariance matrix representation (LCMR) method. Third, sample pairs with the smallest and largest similarities are extracted according to the two distances. Then, these pairs are fed into the siamese multilayer perceptron (MLP) network and discriminative embeddings are learned by training the network with contrastive loss. Finally, density-based spectral clustering is applied to the deep embedding to obtain clustering results. Experimental results on three real HSIs demonstrate that the proposed method can achieve better performance than the considered baseline methods. [ABSTRACT FROM AUTHOR]

We present an algorithm for simultaneous face detection, landmarks localization, pose estimation and gender recognition using deep convolutional neural networks (CNN). The proposed method called, HyperFace, fuses the intermediate layers of a deep CNN using a separate CNN followed by a multi-task learning algorithm that operates on the fused features. It exploits the synergy among the tasks which boosts up their individual performances. Additionally, we propose two variants of HyperFace: (1) HyperFace-ResNet that builds on the ResNet-101 model and achieves significant improvement in performance, and (2) Fast-HyperFace that uses a high recall fast face detector for generating region proposals to improve the speed of the algorithm. Extensive experiments show that the proposed models are able to capture both global and local information in faces and performs significantly better than many competitive algorithms for each of these four tasks. [ABSTRACT FROM AUTHOR]

Objective: A new image instance segmentation method is proposed to segment individual glands (instances) in colon histology images. This process is challenging since the glands not only need to be segmented from a complex background, they must also be individually identified. Methods: We leverage the idea of image-to-image prediction in recent deep learning by designing an algorithm that automatically exploits and fuses complex multichannel information—regional, location, and boundary cues—in gland histology images. Our proposed algorithm, a deep multichannel framework, alleviates heavy feature design due to the use of convolutional neural networks and is able to meet multifarious requirements by altering channels. Results: Compared with methods reported in the 2015 MICCAI Gland Segmentation Challenge and other currently prevalent instance segmentation methods, we observe state-of-the-art results based on the evaluation metrics. Conclusion: The proposed deep multichannel algorithm is an effective method for gland instance segmentation. Significance: The generalization ability of our model not only enable the algorithm to solve gland instance segmentation problems, but the channel is also alternative that can be replaced for a specific task. [ABSTRACT FROM PUBLISHER]

Robust brain magnetic resonance (MR) segmentation algorithms are critical to analyze tissues and diagnose tumor and edema in a quantitative way. In this study, we present a new tissue segmentation algorithm that segments brain MR images into tumor, edema, white matter (WM), gray matter (GM), and cerebrospinal fluid (CSF). The detection of the healthy tissues is performed simultaneously with the diseased tissues because examining the change caused by the spread of tumor and edema on healthy tissues is very important for treatment planning. We used T1, T2, and FLAIR MR images of 20 subjects suffering from glial tumor. We developed an algorithm for stripping the skull before the segmentation process. The segmentation is performed using self-organizing map (SOM) that is trained with unsupervised learning algorithm and fine-tuned with learning vector quantization (LVQ). Unlike other studies, we developed an algorithm for clustering the SOM instead of using an additional network. Input feature vector is constructed with the features obtained from stationary wavelet transform (SWT) coefficients. The results showed that average dice similarity indexes are 91% for WM, 87% for GM, 96% for CSF, 61% for tumor, and 77% for edema. [ABSTRACT FROM PUBLISHER]

Surface electromyography (sEMG) is widely used in evaluating the functional status of the hand to assist in hand gesture recognition, prosthetics and rehabilitation applications. The sEMG is a noninvasive, easy to record signal of superficial muscles from the skin surface. Considering the nonstationary characteristics of sEMG, recent feature selection of hand gesture recognition using sEMG signals necessitate designers to use nonnegative matrix factorization (NMF)-based methods. This method exploits both the additive and sparse nature of signals by extracting accurate and reliable measurements of sEMG features using a minimum number of sensors. The testing has been conducted for simple and complex finger flexions using several experiments with artificial neural network classification scheme. It is shown, both by simulation and experimental studies, that the proposed algorithm is able to classify ten finger flexions (five simple and five complex finger flexions) recorded from two sEMG sensors up to 92% (95% for simple and 87% for complex flexions) accuracy. The recognition performances of simple and complex finger flexions are also validated with NMF permutation matrix analysis. [ABSTRACT FROM PUBLISHER]

Tremor is defined as the involuntary rhythmic or quasi-rhythmic oscillation of a body part, resulting from alternating or simultaneous contractions of antagonistic muscle groups. While tremor may be physiological, those who have disabling pathological tremors find that performing typical activities for daily living to be physically challenging and emotionally draining. Detecting the presence of tremor and its proper identification are crucial in prescribing the appropriate therapy to lessen its deleterious physical, emotional, psychological, and social impact. While diagnosis relies heavily on clinical evaluation, pattern analysis of surface electromyogram (sEMG) signals can be a useful diagnostic aid for an objective identification of tremor types. Using sEMG system attached to several parts of the patient’s body while performing several tasks, this research aims to develop a classifier system that automates the process of tremor types recognition. Finding the optimal model and its corresponding parameters is not a straightforward process. The resulting workflow, however, provides valuable information in understanding the interplay and impact of the different features and their parameters to the behavior and performance of the classifier system. The resulting model analysis helps identify the necessary locations for the placement of sEMG electrodes and relevant features that have significant impact in the process of classification. These information can help clinicians in streamlining the process of diagnosis without sacrificing its accuracy. [ABSTRACT FROM AUTHOR]

We present an algorithm for the online detection of rotor bar breakage in induction motors through the use of wavelet packet decomposition (WPD) and neural networks. The system provides a feature representation of multiple frequency resolutions for faulty modes and accurately differentiates between healthy and faulty conditions, and its main applicability is to dynamic time-variant signals experienced in induction motors during run time. The algorithm analyzes rotor bar faults by WPD of the induction motor stator current. The extracted features with different frequency resolutions, together with the slip speed, are then used by a neural network trained for the detection of faults. The experimental results show that the proposed method is able to detect the faulty conditions with high accuracy. [ABSTRACT FROM AUTHOR]

A maritime automatic target recognition system is developed that performs ship classification using one-dimensional high resolution range profiles. Five physically based features are defined and are extracted from both VV and HH polarizations resulting in a 10-dimensional feature vector. A nonlinear classifier combination approach involving a neural network combiner along with three individual classifiers (Bayes, nearest neighbor, and neural network) is proposed. A decision confidence measure based on the classifier discriminants is developed using a nonparametric estimation approach. The confidence measure enables the system to reject samples that have a low decision confidence. The performance of the proposed neural network based combination is compared with individual classifiers and a number of other combination rules. The results show that this combination can provide high recognition accuracy along with a high probability of declaration. The performance in the presence of samples from not-before-seen classes is also investigated. A new nearest neighbor confidence thresholding approach is developed to aid the neural network combiner in rejecting these samples. [ABSTRACT FROM AUTHOR]