Affect Detection Performance Research Articles

Vine copula Granger causality in mean

Ever since the Granger causality test was proposed in 1969, financial market researchers have used it heavily to determine whether the past of a one-time series facilitates the future prediction of another time series. However, as many researchers have noted, the traditional Granger causality test based on the vector autoregression model has limitations in detecting nonlinear causality. To relax the parametric model assumptions of the Granger causality test, nonparametric versions have been proposed to use the advantage of detecting nonlinear Granger causality but have shown difficulty in selecting smoothing parameters that significantly affect detection performance. To overcome the difficulties of both parametric and nonparametric Granger causality tests, we propose the vine copula Granger causality test in mean based on the semiparametric time-series modeling technique. The proposed test overcomes the shortcomings of parametric modeling and has a computational advantage over the nonparametric tests. Our test shows good size and power performance with various simulated data and a real data.

Multilevel Feature Learning Method for Accurate Interictal Epileptiform Spike Detection.

Interictal epileptiform spike (referred to as spike) detected from electroencephalograms lasting only 20- to 200-ms can provide a reliable evidence-based indicator for clinical seizure type diagnosis. Recent feature representation approaches focus either on the concrete-level or abstract-level information mining of the spike, thus demonstrating suboptimal detection performance. Additionally, existing abstract-level information mining methods of the spike based deep learning networks have not realized the effective feature representation of long-term dependent distinguished information within similar waveform cycles caused by morphological heterogeneity, which affects detection performance. Thus, a multilevel feature learning method for accurate spike detection was proposed in this study. Specifically, the spatio-temporal-frequency multidomain information in concrete-level first are inferred the common mimetic properties of the spike using the multidomain feature extractors. Then, the effective feature representation of long-term dependent distinguished information within similar waveform cycles caused by morphological heterogeneity is suitably captured using the temporal convolutional network. Finally, the spatio-temporal-frequency multidomain long-term dependent feature representation of spike is calculated using the element-wise manner to fuse the feature representation in concrete- and abstract-levels. The experimental results indicate that the proposed method can achieve an accuracy of 90.62±1.38%, sensitivity of 90.38±1.52%, specificity of 91.00±1.60%, precision of 90.33±4.71%, and the false detection rate per minute is [Formula: see text], which are higher than when using the feature representation in the concrete-or abstract-level alone. Additionally, the detection results indicate that the proposed method avoids the subjectivity and inefficiency of visual inspection, and it enables a highly accurate detection of the spike.

Priori Information-Based Feature Extraction Method for Small Target Detection in Sea Clutter

Under the framework of feature-based detection of small targets on sea surface, existing feature extraction methods only use the echo data of current frame while ignoring the influence of historical echo data. Nevertheless, due to the non-stationarity of sea clutter, it may lead to unstable extraction of detection features, and then affect detection performance. To solve this problem, this paper designs a feature extraction method based on <i>a priori</i> information for small target detection. It firstly obtains <i>a priori</i> information from historical echo data by kernel density estimation (KDE) method. Then, the corresponding feature estimation method is utilized to obtain improved feature according to the relationship between current frame data and <i>a priori</i> information. Finally, the feature information of current frame is integrated into <i>a priori</i> information to prepare next feature extraction. Measured data are utilized to verify the performance of proposed method and the results reveal that, this method can effectively improve detection performance especially when sea clutter and target echo have good separability. In addition, the complexity of algorithm is analyzed to prove that proposed method has certain application potential.

Design of QAM-OFDM Radar-Communication Integrated Signal Based on Golay Complementary sequences

The integration of radar and communication is conducive to improving the utilization rate of hardware resources and spectrum. To address the issue that the communication information in orthogonal frequency division multiplexing (OFDM) integrated signal for radar and communication affects detection performance, an OFDM integrated signal for radar and communication based on precoding was proposed. The communication information of the integrated signal was pre-coded by using the Gray Complementary Sequences (GCS) with better correlation performance to improve the performance of the ambiguity function of the integrated signal and its detection ability. Simulation results show that GCS have better correlation performance compared to M and Gold sequences, and the OFDM integrated signal pre-coded by GCS has lower Peak Sidelobes Ratio (PSLR) and comparable radar performance to LFM.

Investigation on method of reducing surface leakage current of the CdZnTe photoconductive detector with MSM structure

The surface leakage current (SLC) of CdZnTe is an important factor affecting detection performance. In this work, we established a two-dimensional model of metal-semiconductor-metal (MSM) structured CdZnTe with a low-resistance layer on the surface. The composition of SLC and the distribution of the internal electric field are analyzed by simulation. The study found that the change of electrode size will change the contribution of SLC from cathode, anode and side surfaces. With the increase of electrode size, the contribution of SLC to total leakage current decreases gradually, and the internal electric field distribution becomes more uniform. According to the conclusions of simulation, we have proposed a simple and accurate method to estimate the bulk resistivity and SLC of CdZnTe. In addition, combined with a series of experiments, we have also summarized several methods which can reduce the contribution of the SLC in MSM structured CdZnTe detector in order to improve energy resolution.

Comparison of Passive Sonar Narrowband Detection and Wideband Detection Performance

Conventional passive sonar uses wideband beam energy for target detection. In the case of low signal-to-noise ratio and strong target interference, the performance of wideband energy detection is rapidly degraded. In view of the relatively stable line spectrum of the radiated noise of underwater targets, the performance of narrowband detection and wideband detection is considered. In stable line spectrum underwater target detection, narrowband detection has performance advantages over wideband detection. However, its actual performance has disadvantages such as sensitivity to noise frequency bands. In this paper, based on the sonar equation, the theoretical methods of wideband detection and narrowband detection are analyzed and the main factors affecting detection performance are given. Simulation and sea trial data processing results verify the effectiveness of the method.

Modeling of artifacts in the wrist photoplethysmogram: Application to the detection of life-threatening arrhythmias

ObjectiveA model for simulating motion-induced artifacts in the wrist photoplethysmogram (PPG) is proposed for the purpose to improve realism of PPG models. MethodsThe database of day-long PPGs, acquired during cardiac rehabilitation, is used to extract artifact characteristics, which further serve as a basis for modeling artifacts in simulated PPGs with life-threatening arrhythmias. ResultsDepending on the recording, 14–49% of the PPG duration is corrupted by artifacts, mostly due to device displacement, forearm and hand motion. The artifact type influence on the performance of a life-threatening arrhythmia detector shows that the sensitivity drops by 45–48% for extreme bradycardia and by 13–32% for ventricular tachycardia. Poor contact causes 2–4 times more false alarms of ventricular tachycardia compared to the other artifact types under investigation. ConclusionSimulation of realistic artifacts encountered in activities of daily living allows to comprehensively investigate arrhythmia detectors and understand the artifact types most negatively affecting detection performance. SignificanceThe proposed PPG artifact model is of importance for developing and testing artifact-robust arrhythmia detectors.

Developing and Evaluating Poultry Preening Behavior Detectors via Mask Region-Based Convolutional Neural Network.

Simple SummaryPreening is poultry grooming and comfort behavior to keep plumages in good conditions. Automated tools to continuously monitor poultry preening behaviors remain to be developed. We developed and evaluated hen preening behavior detectors using a mask region-based convolutional neural network (mask R-CNN). Thirty Hy-line brown hens kept in an experimental pen were used for the detector development. Different backbone architectures and hyperparameters (e.g., pre-trained weights, image resizers, etc.) were evaluated to determine the optimal ones for detecting hen preening behaviors. A total of 1700 images containing 12,014 preening hens were used for model training, validation and testing. Our results show that the final performance of detecting hen preening was over 80% for precision, recall, specificity, accuracy, F1 score and average precision, indicating decent detection performance. The mean intersection over union (MIOU) was 83.6–88.7%, which shows great potential for segmenting objects of concern. The detectors with different architectures and hyperparameters performed differently for detecting preening birds and thus we need to carefully adjust these parameters to obtain a robust deep learning detector. In summary, deep learning techniques may have a great ability to automatically monitor poultry behaviors and assist welfare-oriented poultry management.There is a lack of precision tools for automated poultry preening monitoring. The objective of this study was to develop poultry preening behavior detectors using mask R-CNN. Thirty 38-week brown hens were kept in an experimental pen. A surveillance system was installed above the pen to record images for developing the behavior detectors. The results show that the mask R-CNN had 87.2 ± 1.0% MIOU, 85.1 ± 2.8% precision, 88.1 ± 3.1% recall, 95.8 ± 1.0% specificity, 94.2 ± 0.6% accuracy, 86.5 ± 1.3% F1 score, 84.3 ± 2.8% average precision and 380.1 ± 13.6 ms·image−1 processing speed. The six ResNets (ResNet18-ResNet1000) had disadvantages and advantages in different aspects of detection performance. Training parts of the complex network and transferring some pre-trained weights from the detectors pre-trained in other datasets can save training time but did not compromise detection performance and various datasets can result in different transfer learning efficiencies. Resizing and padding input images to different sizes did not affect detection performance of the detectors. The detectors performed similarly within 100–500 region proposals. Temporal and spatial preening behaviors of individual hens were characterized using the trained detector. In sum, the mask R-CNN preening behavior detector could be a useful tool to automatically identify preening behaviors of individual hens in group settings.

Group Sampling for Scale Invariant Face Detection.

Detectors based on deep learning tend to detect multi-scale objects on a single input image for efficiency. Recent works, such as FPN and SSD, generally use feature maps from multiple layers with different spatial resolutions to detect objects at different scales, e.g., high-resolution feature maps for small objects. However, we find that objects at all scales can also be well detected with features from a single layer of the network. In this paper, we carefully examine the factors affecting detection performance across a large range of scales, and conclude that the balance of training samples, including both positive and negative ones, at different scales is the key. We propose a group sampling method which divides the anchors into several groups according to the scale, and ensure that the number of samples for each group is the same during training. Our approach using only one single layer of FPN as features is able to advance the state-of-the-arts. Comprehensive analysis and extensive experiments have been conducted to show the effectiveness of the proposed method. Moreover, we show that our approach is favorably applicable to other tasks, such as object detection on COCO dataset, and to other detection pipelines, such as YOLOv3, SSD and R-FCN. Our approach, evaluated on face detection benchmarks including FDDB and WIDER FACE datasets, achieves state-of-the-art results without bells and whistles.

Opinion Spam Detection based on Annotation Extension and Neural Networks

Online reviews play an increasingly important role in the purchase decisions of potential customers. Incidentally, driven by the desire to gain profit or publicity, spammers may be hired to write fake reviews and promote or demote the reputation of products or services. Correspondingly, opinion spam detection has attracted attention from both business and research communities in recent years. However, unlike other tasks such as news classification or blog classification, the existing review spam datasets are typically limited due to the expensiveness of human annotation, which may further affect detection performance even if excellent classifiers have been developed. We propose a novel approach in this paper to boost opinion spam detection performance by fully utilizing the existing labelled small-size dataset. We first design an annotation extension scheme that uses extra tree classifiers to train multiple estimators and then iteratively generate reliable labelled samples from unlabeled ones. Subsequently, we examine neural network scenarios on a newly extended dataset to learn the distributed representation. Experimental results suggest that the proposed approach has better generalization capability and improved performance than state-of-the-art methods.

Engine Misfire Diagnosis Based on the Torsional Vibration of the Flexible Coupling in a Diesel Generator Set: Simulation and Experiment

Flexible couplings are widely used for accommodating various imprecisions such as misalignments and torque fluctuations in a diesel generator set. However, excessive vibration due to common engine misfires can cause premature failures in both the couplings and the whole generator set. Moreover, the coupling also affects detection performance and the occurrences of early misfires due to such faults in fuel injections and cylinder seals. In this paper, the torsional vibration along with the dynamic torque of the flexible coupling connecting a 6-cylinder diesel engine to a dynamometer is studied theoretically and experimentally. This system represents a common diesel generator set which is modeled into a torsional dynamic system with 10-degree of freedom (DOF) by the Lagrange equation method. After a verification of the model by the measurement of vibration responses at the free end and the coupling with a magnetic speed sensor and a wireless torque sensor, it is then used to predict the influences of different misfire conditions on the dynamic torque on the coupling, which helps in determining reliable diagnostic features for misfire detection and diagnosis. Results show that the misfires of the diesel engine have significant influences (more than 5 times higher) on the torsional vibration and load of the flexible coupling than on the dynamic torque. Nevertheless, both the torsional vibration and torque of the flexible coupling can be used to detect the occurrence of the diesel engine misfires and the vibration provides slight better results in terms of localizing the misfired cylinders.

Enhancing Multi-Camera People Detection by Online Automatic Parametrization Using Detection Transfer and Self-Correlation Maximization †

Finding optimal parametrizations for people detectors is a complicated task due to the large number of parameters and the high variability of application scenarios. In this paper, we propose a framework to adapt and improve any detector automatically in multi-camera scenarios where people are observed from various viewpoints. By accurately transferring detector results between camera viewpoints and by self-correlating these transferred results, the best configuration (in this paper, the detection threshold) for each detector-viewpoint pair is identified online without requiring any additional manually-labeled ground truth apart from the offline training of the detection model. Such a configuration consists of establishing the confidence detection threshold present in every people detector, which is a critical parameter affecting detection performance. The experimental results demonstrate that the proposed framework improves the performance of four different state-of-the-art detectors (DPM , ACF, faster R-CNN, and YOLO9000) whose Optimal Fixed Thresholds (OFTs) have been determined and fixed during training time using standard datasets.

Abandoned Object Detection in Video-Surveillance: Survey and Comparison.

During the last few years, abandoned object detection has emerged as a hot topic in the video-surveillance community. As a consequence, a myriad of systems has been proposed for automatic monitoring of public and private places, while addressing several challenges affecting detection performance. Due to the complexity of these systems, researchers often address independently the different analysis stages such as foreground segmentation, stationary object detection, and abandonment validation. Despite the improvements achieved for each stage, the advances are rarely applied to the full pipeline, and therefore, the impact of each stage of improvement on the overall system performance has not been studied. In this paper, we formalize the framework employed by systems for abandoned object detection and provide an extensive review of state-of-the-art approaches for each stage. We also build a multi-configuration system allowing one to select a range of alternatives for each stage with the objective of determining the combination achieving the best performance. This multi-configuration is made available online to the research community. We perform an extensive evaluation by gathering a heterogeneous dataset from existing data. Such a dataset allows considering multiple and different scenarios, whereas presenting various challenges such as illumination changes, shadows, and a high density of moving objects, unlike existing literature focusing on a few sequences. The experimental results identify the most effective configurations and highlight design choices favoring robustness to errors. Moreover, we validated such an optimal configuration on additional datasets not previously considered. We conclude the paper by discussing open research challenges arising from the experimental comparison.

Research on the optimal optical attenuation in a laser radar using a Geiger-mode APD

For a laser radar (LADAR) system using a Geiger-mode avalanche photodiode (GmAPD), attenuating echo and background noise simultaneously affect the original data output from the GmAPD and eventually affect detection performance. In this study, we established a model that applies to the GmAPD-based LADAR with optical attenuation and also applies to any typical single photon detector that has a dead time (e.g.,the photomultiplier tube); thus, a comprehensive and fundamental study is performed for the mathematical expectation of the number of signal detections (ES), the mathematical expectation of the number of noise detections (EN), the signal-to-noise ratio (SNR), and the range bias (absolute error, Rb) and precision (standard deviation, Rp) under various attenuation levels with different dead times and signal noise conditions. We observed the following: on the one hand, there exists an optimum attenuation level at which ES and SNR are maximized; on the other hand, there exists another optimum attenuation level for shorter dead times, at which Rp is minimized. The phenomenon of the maximum ES, SNR, or minimum Rp disappears gradually as the echo or noise decreases from high levels (e.g.,10 photoelectrons/echo or an equivalent background noise of 10 photoelectrons/range gate). Further, higher attenuation, which shows advantages under strong echo or noise conditions, yields a larger improvement in ES for longer dead times; and, with the reduction of the dead time or the noise, the maximum ES gradually increases, and the corresponding optimum attenuation level becomes slighter. Additionally, we found that, as the optical attenuation increases, EN decreases to 0, Rb changes from a negative value to 0, and Rp is minimized, becomes slightly worse, and reaches a constant. Moreover, the shorter dead times, which show advantages when they are shorter than the end time of the echo, lead to a larger ES, better Rb, and slightly worse Rp than the longer ones.

A Trainable System for Underwater Pipe Detection

Underwater image processing is widely increased over the last decade. It is a fundamental process for a most part of underwater research applications, because of the need of data acquisition. In this paper we will propose a novel approach of pipe detection in submarine environment. The system draws much of its power from a representation that describes an object class taking into account structure and content features which are computed through the multi-scale covariance descriptor. This approach describes an object detection model by training a support vector machine classifier using a large set of positive and negative samples. We present result on pipe detection using Maris dataset. Moreover, we show how the representation affects detection performance by considering mono-scale representation using Covariance descriptor.

Manifold learning techniques for unsupervised anomaly detection

Abstract Appropriately identifying outlier data is a critical requirement in the decision-making process of many expert and intelligent systems deployed in a variety of fields including finance, medicine, and defense. Classical outlier detection schemes typically rely on the assumption that normal/background data of interest are distributed according to an assumed statistical model and search for data that deviate from that assumption. However, it is frequently the case that performance is reduced because the underlying distribution does not follow the assumed model. Manifold learning techniques offer improved performance by learning better models of the background but can be too computationally expensive due to the need to calculate a distance measure between all data points. Here, we study a general framework that allows manifold learning techniques to be used for unsupervised anomaly detection by reducing computational expense via a uniform random sampling of a small fraction of the data. A background manifold is learned from the sample and then an out-of-sample extension is used to project unsampled data into the learned manifold space and construct an anomaly detection statistic based on the prediction error of the learned manifold. The method works well for unsupervised anomaly detection because, by definition, the ratio of anomalous to non-anomalous data points is small and the sampling will be dominated by background points. However, a variety of parameters that affect detection performance are introduced so we use here a low-dimensional toy problem to investigate their effect on the performance of four learning algorithms (kernel PCA, two versions of diffusion map, and the Parzen density estimator). We then apply the methods to the detection of watercraft in an ensemble of 22 infrared maritime scenes where we find kernel PCA to be superior and show that it outperforms a commonly employed baseline algorithm. The framework is not limited to the tested image processing example and can be used for any unsupervised anomaly detection task.

Personalised models for speech detection from body movements using transductive parameter transfer

We investigate the task of detecting speakers in crowded environments using a single body worn triaxial accelerometer. Detection of such behaviour is very challenging to model as people’s body movements during speech vary greatly. Similar to previous studies, by assuming that body movements are indicative of speech, we show experimentally, on a real-world dataset of 3 h including 18 people, that transductive parameter transfer learning (Zen et al. in Proceedings of the 16th international conference on multimodal interaction. ACM, 2014) can better model individual differences in speaking behaviour, significantly improving on the state-of-the-art performance. We also discuss the challenges introduced by the in-the-wild nature of our dataset and experimentally show how they affect detection performance. We strengthen the need for an adaptive approach by comparing the speech detection problem to a more traditional activity (i.e. walking). We provide an analysis of the transfer by considering different source sets which provides a deeper investigation of the nature of both speech and body movements, in the context of transfer learning.

Privacy-Preserving Fraud Detection Across Multiple Phone Record Databases

Subscription fraud, i.e., customers signing up to a service with no intent to pay, causes significant losses in the telecommunication industry. Telecom operators have developed strategies to identify those fraudsters, but fraudsters tend to migrate from one carrier to another. Data sharing between telecoms would increase fraud detection rates, but phone records are protected by law and telecom operators might be reluctant to share information about fraudsters because they see it as giving a competitive advantage. We propose several protocols to enable fraud detection across multiple databases without revealing additional information. We also propose a model to generate phone records, with which we evaluate how the choice of parameters affects detection performance. We show feasibility, performance and costs with implementations of our protocols.

One dimensional semiconductor nanostructures: An effective active-material for terahertz detection

One-dimensional (1D) nanostructure devices are at the frontline of studies on future electronics, although issues like massive parallelization, doping control, surface effects, and compatibility with silicon industrial requirements are still open challenges. The recent progresses in atomic to nanometer scale control of materials morphology, size, and composition including the growth of axial, radial, and branched nanowire (NW)-based heterostructures make the NW an ideal building block for implementing rectifying diodes or detectors that could be well operated into the Terahertz (THz), thanks to their typical achievable attofarad-order capacitance. Here, we report on our recent progresses in the development of 1D InAs or InAs/InSb NW-based field effect transistors exploiting novel morphologies and/or material combinations effective for addressing the goal of a semiconductor plasma-wave THz detector array technology. Through a critical review of material-related parameters (NW doping concentration, geometry, and/or material choice) and antenna-related issues, here we underline the crucial aspects that can affect detection performance across the THz frequency region.

Characterization of crosstalk in stereoscopic display devices.

Many different types of stereoscopic display devices are used for commercial and research applications. Stereoscopic displays offer the potential to improve performance in detection tasks for medical imaging diagnostic systems. Due to the variety of stereoscopic display technologies, it remains unclear how these compare with each other for detection and estimation tasks. Different stereo devices have different performance trade‐offs due to their display characteristics. Among them, crosstalk is known to affect observer perception of 3D content and might affect detection performance. We measured and report the detailed luminance output and crosstalk characteristics for three different types of stereoscopic display devices. We recorded the effect of other issues on recorded luminance profiles such as viewing angle, use of different eye wear, and screen location. Our results show that the crosstalk signature for viewing 3D content can vary considerably when using different types of 3D glasses for active stereo displays. We also show that significant differences are present in crosstalk signatures when varying the viewing angle from 0 degrees to 20 degrees for a stereo mirror 3D display device. Our detailed characterization can help emulate the effect of crosstalk in conducting computational observer image quality assessment evaluations that minimize costly and time‐consuming human reader studies.