This PDF file contains the front matter associated with Volume 11384, including the Title Page, Copyright information, Table of Contents, Introduction and the Conference Committee list

Clutter suppression, especially in time-varying environments is a hindrance that must be solved for radar systems applied to unmanned vehicles. However, exponential moving average (EMA) method, a common background subtraction technique, does not handle such a situation very well because the fixed parameter constrains the updating of the estimated clutter. In this paper, we propose a novel adaptive clutter suppression algorithm to adjust the parameter of EMA method under the background of time-varying clutter. The main idea is to adopt a low-complexity time-averaged variable forgetting factor (TAVFF) mechanism. The proposed algorithm is assessed with data recording measured background clutter and a simulated moving target. The simulation results demonstrate our proposed algorithm has achieved both fast convergence and good steady-state performance.

Radar signal sorting is the key technology of electronic warfare, and pulse repetition interval (PRI) is an important parameter of signal sorting. In this paper, a PRI sinusoidal extraction method of modulation feature based on Empirical mode decomposition (EMD) decomposition is proposed. By defining the S function and performing EMD decomposition on it, the Intrinsic Mode Function (IMF) group obtained. Selecting the appropriate IMF component to extract the sinusoidal modulation period. Combined with the S function, the pulse sequence initially screened, and the modulation characteristics are determined according to the screening results. A pulse sorting algorithm is implemented according to the modulation characteristics. The simulation results show that the proposed method can effectively extract the modulation information from multiple radar pulses with different modulation periods, such as the modulation period of the PRI modulated signal, and complete the sorting of the radar pulse.

Aiming at the problem of Direction of Arrival (DOA) tracking for multiple target, this paper proposes a DOA tracking algorithm based on Propagator Method (PM) under Multi-Bernoulli filtering framework. The proposed algorithm uses particle filter to approximate the posterior distribution of target, where the calculation of likelihood function is the key of the update step. The eigendecomposition of the covariance matrix is needed when the likelihood function is replaced by MUSIC spatial spectrum function. In order to reduce the computational complexity of the matrix eigendecomposition, we use the spatial spectral function of PM to replace the pseudo-likelihood function of particle filter, and further exponential weighting is used to enhance the weight of particles at high likelihood area and make resampling more efficient. The simulation results show that the proposed algorithm can effectively track the DOA and estimate the number of multiple maneuvering target.

In the X-ray pulsar navigation process, since the pulsar signal obtained by the epoch folding contains a large amount of noise, the signal must be denoised in order to obtain higher positioning accuracy. In order to further optimize the denoising effect and improve the algorithm in real time, this paper proposes a pulsar wavelet base and implements its lifting scheme. In this paper, wavelet multi-level decomposition is performed on the pulsar outline, then a wavelet base based on the pulsar's own signal is constructed according to the low-frequency coefficients, and its lifting method is realized. Matlab simulation shows that compared with db4 and db5 methods, the proposed method performs better in terms of signal-to-noise ratio, mean square error, peak relative error, peak position error and real-time performance. Although the peak error of the db1 wavelet is relatively small, its signal-to-noise ratio is too large, and the overall performance is obviously not as good as the proposed method. The proposed signal-to-noise ratio is up to 4.2dB higher than the db4 and db5 methods, and the mean square error is only 24.3% of the db4 and db5 methods. The peak position error is only 50% of the db4 and db5 methods.

Achieving stationary speech enhancement in low signal-to-noise ratio (SNR) environments is a challenging problem. Because noise energy is dominant in noisy speech at low SNR level, the existence of numerous obvious random noises may lead neural network to forget some useful information obtained by early training. Moreover, it is difficult for a single neural network to obtain effective speech features and noise features. Therefore, this paper designs to utilize multiple neural networks in two stages to discriminately learn a certain type of noise features and reduce the introduction of interference. Experiment results demonstrate that proposed method leads to consistently better source-to-distortion ratio (SDR) and perceptual evaluation of speech quality (PESQ) than baseline models in low SNR condition. And the results indicate that the method can suppress the forgetting of early information of neural network.

In the question of the impact of the main-beam width on vessel radiated noise power spectrum characteristics, a typical mathematical model was established, the influence of the main-beam covering vessel and its trajectory on the acquisition of vessel radiation noise power spectrum characteristics is studied. Theoretical and simulation results show that when main-beam cannot cover the vessel and its trajectory, the method of multi-beam synthesis should be apply in order to undistorted acquire the vessel radiation noise power spectrum characteristic. A cargo-ship radiated noise characteristics is acquired based on the method of multi-beam synthesis in this paper, the difference of broadband sound pressure level acquired by beamforming and single hydrophone is less than 1.8dB. The research results can provide a basis and example for the effective acquisition of vessel radiated noise characteristics based on acoustic measurement array.

Angle Random Walk (ARW) is an important performance index of the MEMS gyroscope, which is determined by the noise in the control system. In this paper, the thermal noise of gyroscope, the noise of pickoff circuit and feedback circuit are considered, and the noise model of control system is constructed under force-to-rebalance (FTR) closed-loop detection. The influence of system parameters on the power spectral density (PSD) of noise equivalent rate (NER) is analyzed, and the noise is reduced by adjusting system parameters reasonably. Simulink numerical simulation proves the correctness of the noise model.

Interrupted sampling repeater jamming (ISRJ), which is generated by digital radio frequency memory (DRFM), has become a hot topic about electronic counter measures (ECM). The paper has proposed an ISRJ suppression algorithm based on numerical statistical characteristic analysis and clustering. First, preprocess the echo data to improve signalnoise ratio according to the numerical statistical characteristic analysis. Next, apply the algorithm of clustering to the echo so as to identify ISRJ and obtain the parameters of ISRJ. Then, reconstruct the jamming based on the ISRJ information from previous step. Finally, use the reconstructed jamming to suppress ISRJ. The simulation result shows that the proposed method can identify ISRJ successfully when SNR of echo is higher than -10dB.

Multipath interference is the main threat to the ultra-low targets detection. A novel method of suppressing multipath interference based on Brewster Effect is proposed. The traditional Four-Path Model is modified by complex reflection coefficient and antenna pattern. The numerical hybrid method PO+MEC is used to calculate the scattering fields of targets. The method based on scattering center model is introduced to generate the echo signal. The effect of the method proposed in this paper is analyzed and proved in two aspects, scattering field and echo signal. The conclusion is that, if the active seeker detect the ultra-low target using the Brewster angle as the grazing angle in VV-polarization, the multipath interference is well suppressed. Under maritime environment, the Brewster angle is approximate 7 . The work in this paper is of great significance in military field.

In frequency domain, moving targets with different velocities will affect the different frequency components of sea clutter spectrum. In order to analyze the nonlinear influence, the piecewise fractional Brownian motion is introduced, which builds the relation between different spectrum parts and the fine/coarse scales, i.e. the target echoes at different velocities can affect the self-similarities of sea clutter at different scales in time domain. Based on the real X-band and Sband sea clutter data, this influence mechanism is studied and verified. The results show that the slow-moving target mainly affects the self - similarity of the sea clutter sequence at the coarse time scales, and the fast-moving target mainly affects the self - similarity of the sea clutter sequence at the fine time scales. This conclusion lays the foundation for introducing the fractal theory into the sea clutter spectrum analysis and target detection in frequency domain.

In order to improve the accuracy of on-line dynamic reconstruction of time-varying signals, a dynamic compressed sensing algorithm based on sparse Bayesian learning named Support-DCS is proposed in this paper. Since there is no need to assume any time-varying law of signal and no need to adjust any model parameters artificially, the algorithm has good adaptability. Three time-varying signal types with different correlation levels are set up for experiments. The experimental results showed that, compared with several main existing algorithms, the proposed algorithm always has remarkable advantages in signal-to-error ratio, while the other algorithms can only be effective when the support set changes slowly and the signal amplitude correlation is strong.

This paper proposes a note segmentation method combining music score, aiming at the problem that cannot segment multi-tone music with more changes in intensity accurately. By extracting the envelope peak value of music signal and matching it with note values and pitch information of musical score, the segmentation of notes is completed.Simulation results show that the method based on the prior knowledge of value and pitch information of music score can not only realize the segmentation of notes of continuous single tone music, but also be suitable for multi-tone music with strong and weak variations.

The importance of Unmanned Arial Vehicle (UAV) has made progressive usage in recent times due to ease of availability and miniaturization. While on another hand, it might pose a malicious effect on public safety, so the most important problem to be addressed is the recognition of drones in sensitive areas. This paper addressed the machine learning approach to recognize UAV through its acoustic emission using representative algorithms of Mel frequency cepstral coefficients (MFCCs) for feature extraction and random forest (RNF) classifier for classification. However, temporal and spectral features are devised to demonstrate performances of beam-formed signals (enhanced emitter at desired direction) and raw signal (captured in flying test). Results of extracted features from a beam-formed signal, demonstrate the effectiveness of MFCC performance regardless of a noisy environment with a high accuracy rate as compared to raw signal. RNF classifier was trained to classify feature vector, which is obtained from the feature extraction stage. However, the classifier helped to classify samples from a small data set with good accuracy. It can appropriately classify with a likelihood of around 75% under various training data sets.

In the process of extracting rotor features using time-frequency analysis, clutter suppression and improving time-frequency resolution have always been problems that need to be solved and improved. The paper proposes a rotor feature extraction method with high time-frequency resolution that can suppress clutter. Firstly, the separation of the micro-motion target and the clutter is realized by the complex empirical mode decomposition (CEMD). The high-resolution time-frequency diagram of the rotor is obtained by the synchrosqueezing improved S transform (SIST) proposed in the paper. The features extracted from the diagram are of high accuracy. The simulation results show that this method (CEMD-SIST) has better clutter suppression performance and higher time-frequency resolution than other rotor feature extraction methods.

Speech enhancement is a challenging and critical task in the speech processing research area. In this paper, we propose a novel speech enhancement model based on Wasserstein generative adversarial networks, called WSEM. The proposed model operates on frame-level speech segments by using an adjacent frames extension mechanism, to enforce the mapping from noisy speech to the clean target, which makes it distinctly different from other related GAN-based models. We compare the performance of WSEM with related works on benchmark datasets under different signal-to-noise (SNR) conditions, experimental results show that WSEM performs comparable to the state-of-the-art approaches in all the tests, and it performs especially well in low SNR environments.

The merchant ships including passenger ship target are one of important source of sea environment noise, and is also the primary background jam of underwater passive detection and target recognition, especially areas in shore. Consequently, this paper mainly focuses on measuring and analyzing acoustic spectrum and signature of underwater radiated noise originated from a passenger ship with more than ten thousands of tonnages, especially focusing on the distribution of power and frequency, the amount of stable line-spectrum, structure of envelope modulation spectrum and its stability of amplitude. And the analysis results can be further applied in the studies on underwater passive detection, target recognition and ship acoustic design, especially for passenger ships.

Photoplethysmographic (PPG) signal is an important body sign data, this paper establishes a physiological model by combining linear dynamics method with important physiological variables (mean arterial pressure and heart rate) extracted from photoplethysmographic (PPG), and verifies the relationship between PPG and SIRS: the reduction in the coupling of mean arterial pressure and heart rate characteristics obtained from PPG signals is significantly associated with systemic inflammatory response syndrome(SIRS) symptoms, which remains conspicuous even though after adjusting clinical intervention. Through PPG signal analysis of 270 adult ICU patients from PhysioNet database, power spectrum and transfer function analysis of the method are carried out, and verifies that the method proposed in this paper can be used to reveal the changes associated with SIRS, which provides a possibility for long-term continuous monitoring or detection of SIRS risk for ICU patients under non-invasive conditions.

Aiming at the multi-standard and multi-system requirements of modern wireless communication systems, it is more and more important to process a large number of signals and data efficiently, quickly and relatively low-cost. The software radio technology has wireless communication by its easy modification and low cost. Technology has been upgraded to a new level of software and ease of expansion. Firstly, the GNU Radio platform, software architecture and hardware platform are described. Then, using the general software radio platform, the simulation and implementation of communication channels such as fading and drift are completed in the GNU Radio programming environment. Based on this, GNU Radio+ USRP is further designed. Transmission and reception of modulated signals of GMSK. Thereby wireless communication based on software radio platform is realized. From the experimental results, the system can transmit and receive GMSK modulated signals better.

This study proposes to classify cricket calls using feature selection and ensemble learning in noisy environments. After collecting cricket calls, we first extract both temporal and frequency features from each frame. Then, statistical features over all frames are calculated including mean, variance, skewness, and kurtosis. For temporal feature, we use zero crossing rate, short-time energy and Shannon entropy. Frequency features include Mel-frequency Cepstral coefficients, spectral centroid, spectral entropy, spectral flux, and spectral roll-off. Next, minimum redundancy maximum relevance is used to select important features and remove redundant information. Finally, ensemble learning of four standard classifiers is used to classify cricket call species and types in noisy environments: k-nearest neighbor, logistic regression, Gaussian naïve Bayes, and random forest. Experimental result shows that the best classification F1-score is 89.5% for classifying five cricket species and two cricket types.

In the field of sonar detection, the most commonly used method for direction of arrival (DOA) estimation of underwater targets is the beamforming algorithm. However, due to the Rayleigh limit of resolution, this method cannot effectively resolve multiple targets within one beam. In this paper we propose a DOA estimation method using a single snapshot to resolve two targets in a single beam. We first establish an echo model of two unresolved targets with sonar array. Then we derive an improved monopulse method to estimate the DOA of the targets according to the maximum likelihood estimation principle. Finally, the performance of this method is evaluated by comparison experiments in the cases of varying SNR, inter-target angle separation and inter-target amplitude differences. The simulation results indicate that, method performs very well in many aspects, including smaller estimation error and enhanced adaptation to inter-target amplitude difference.

The purpose of multi-radar echo band fusion is to expand the signal bandwidth and improve the range resolution. Due to the difference of multi-radar space positions and the initial phase of echo, the phase Angle difference is linear phase and fixed phase. Therefore, solving the phase mismatch is the premise and focus of bandwidth fusion. The disadvantage of the existing phase-coherent registration method is that the phase-coherent registration accuracy is low at high noise level. Based on the geometrical diffraction theory (GTD) model, the algorithm analyzes the influence of non-coherent factors on the target echo, and uses the global minimum entropy criterion as the cost function to estimate the linear phase. According to the influence of fixed phase on the target scattering center parameters, a linear least squares method for correcting the pole subspace is proposed to estimate the fixed phase. The simulation results show the advantages and feasibility of the proposed algorithm.

Regarding the main-lobe jamming suppression algorithm of phased array radar, the paper analyzes various anti-jamming algorithms based on blocking matrix pre-processing (BMP) according to the study of current anti-main lobe jamming technique in the space-time domain, including the weighting coefficient compensation, whitening, diagonal loading and linear constraint combined with diagonal loading beam retention algorithms. Inspired by the covariance matrix reconstruction (CMR) algorithm of eigen-projection matrix preprocessing, the paper combines modified CMR and BMP algorithm to suppress main lobe jamming. The modified CMR can not only be used when the dimension is lost caused by BMP, but also solve the distortion problems such as the main lobe peak offset in the adaptive beam forming synthesis. The biggest advantage of BMP combined with modified CMR algorithm is that its anti-jamming performance is excellent and stable when the sampling snapshot contains the target signal. Meanwhile, the algorithm complexity and the snapshot sensitivity are both in a low level. In the end, the verification results of the measured data also show the superiority of the proposed algorithm when the sampling snapshot contains the target signal.

Constant false alarm rate (CFAR) detectors are widely used in modern radar system to declare the presence of targets. Due to the serious masking effects under the multiple targets situation and the clutter edge, the detection probability of CFAR detectors decrease sharply and the alarm rates increase significantly. To solve these problems, a robust adaptive amplitude iteration CFAR (AAI-CFAR) algorithm is proposed in this paper and obtains good performance. By combining the 2nd-order statistic, variability index, and the 4th-order statistic, kurtosis, a variable scaling factor is designed in the amplitude iteration to adapt different environment. Plenty of Monte Carlo simulations are applied to evaluate the performance of the proposed method under different clutter scenarios compared with existing CFAR detectors, which illustrate the superiority and robustness of AAI-CFAR.

Touchless hand gesture recognition is of great importance for human-computer interaction (HCI). In this paper, we present a hand gesture recognition approach based on range-Doppler-angle trajectory and the long short-term memory (LSTM) network with a 77GHz frequency modulated continuous wave (FMCW) multiple-input-multiple-output (MIMO) radar. Firstly, the hand gesture fast-time-slow-time-antenna 3 dimension (3D) data are collected by the FMCW MIMO radar. Additionally, by performing the discretize Fourier transform (DFT) to the fast-time and slow-time, respectively, we obtain the range-profile and Doppler-profile. Then, by using the multiple signal classification (MUSIC) approach, we estimate the angle-profile of the hand gestures. To smooth and eliminate the noise effects, we apply the Kalman filtering to the estimated range-profile, Doppler-profile and angle-profile, respectively, and obtain the range-Doppler-angle trajectory signature. After that, by exploiting the temporal and spatial correlations, we construct a LSTM network for the hand gesture recognition. Experiments with 6 hand gestures are conducted and show that the proposed approach can recognize 6 hand gestures with an average accuracy over 97%.

Recently, hand gesture recognition based-on radar has attracted many researchers in the field of human–computer interfaces. However, the number of kinds of hand gestures and recognition accuracy can be still increased. In this paper, we propose a hand gesture recognition approach based on convolutional neural network (CNN) using a bistatic radar system. Firstly, we build a bistatic radar system which consists of two pulse radars and define an active area of hand gesture called gesture desktop. Then, two time-distance maps are obtained by signal pre-processing, and we build a Bistatic-CNN with two branches of convolutional layers as a classifier to recognize 14 hand gestures. The bistatic radar system can offer us much more information of hand gesture from different perspectives and achieve much higher hand gesture recognition accuracy than single radar. The experimental results based on the measured data show that the proposed approach can recognize 14 hand gestures with average accuracy over 98%.

The performance of radar automatic target recognition (ATR) highly depends on the quality of training database, the extracted features and classification algorithm. Radar target is detected by the Doppler effect in radar echo signal. Through processing the echo signals in different domains, the distinctive characteristic can be obtained intuitively. Furthermore, we can utilize the extracted features to complete radar target classification. This paper proposes a novel target recognition method based on 1D-convolution neural network (CNN) aiming at the ATR of low-resolution ground surveillance radar. The proposed approach uses 1D-CNN as feature extractor and softmax layer as classifier. We tested our method on actual collected database to classify human and car, which reached an accuracy of 98%. Compared with conventional artificial feature extraction approaches, our model shows better performance and adaptability.

A model-based joint tracking and classification (JTC) method is proposed for narrowband radar with kinematic and radar cross section (RCS) measurements. The method is derived from the 3D scattering center model (3DSCM), which can construct an explicit relation between the aspect angle and the predicted RCS. To deal with the numerical problem in observation model, a modified likelihood function for RCS measurement is adopted under the assumption of additive Gaussian observation noise. The JTC processing is realized by sequential Monte Carlo (SMC) technique. Specifically, a bank of particle filters are used to obtain type-dependent target state and type estimates. Compared with the traditional JTC methods using low resolution sensor, the proposed method is free from the constraint that target classification has to rely on different maneuvering modes. Simulation results validate the effectiveness of the proposed method with maritime application scenario.

In this paper, models of jamming signals are established based on the mechanism of active jamming signals against LFM radar. Five time-domain characteristics and frequency-domain characteristics of jamming signals are extracted. The decision tree method, BP neural network method and decision tree support vector machine (DTSVM) method are used to establish the classification models, and the simulation is performed for identifying and classifying the jamming signals at different jamming-to-noise ratio (JNR). The result shows that the model based on DTSVM method has better adaptability, smaller calculation and higher recognition success rate at low JNR.

As an effective method in signal reconstruction model, compressed sensing (CS) has achieved excellent performance in sparse array reconstruction. However, it is necessary to set the penalization factor before iterative calculation, which will increase the difficulty to convergence the result to the global optimal solution. In this paper, we remove the process of choosing penalization factor and reconstruction error by modifying the iterative expression as well as alternating direction method of multipliers (ADMM) algorithm respectively. In addition, the improved model is shown to be convex and thus can be solved using the CVX toolbox. Simulation result shows that the reference pattern could be reconstructed with minimum number of antenna elements by the proposed algorithms. Moreover, the proposed methods have significant performance improvement in main sidelobe level (MSL).

Online reviews are significant sources of information, which is useful for supporting customer and entrepreneur decision in terms of product and service satisfaction analysis. Online reviews containing feedback from various domains makes it difficult to analyze and classify all comments at once. The proposed technique analyses the cross-domain Thai review data using a co-train machine learning model. The co-train model consists of multiple single domain specific models followed by refinement analysis for the final sentiment classification. This allows for full flexibility in training of each individual domain, which can lessen the limitation on training complexity due to simple training on single domain. The experiments have been conducted on Wongnai restaurant domain and IMDB movie domain data. Our co-train model can achieve the highest average accuracy of 86.10 percent for cross-domain sentiment classification with approximately 38 seconds processing time.

When wireless sensor network is used for indoor trajectory tracking of mobile nodes, the random noise in the environment will affect the stability and accuracy of localization. In order to solve this problem, this paper proposed an indoor trajectory tracking algorithm based on Kalman filter and geomagnetic intensity. Firstly, RSSI data are measured by the Zigbee nodes of the wireless network and the Kalman filter algorithm is used to track the trajectory. Then the distance measurement algorithm is used to search and match the geomagnetism data to locate the trajectory further. In order to verify the effectiveness of this algorithm, the indoor moving target will be located in real time through the measured data. The experimental results show that the localization has a better accuracy after adding the results of geomagnetic intensity matching.

This paper proposes the use of RLS real-time high-precision positioning algorithm in high-precision real-time positioning of space-based internet of things loads, which overcomes the traditional least-squares high order of inverse matrix, large amount of calculation and easy to appear ill-conditioned matrix that can’t seek the inverse, can continuously obtaining estimates in real time, suitable for satellite on-orbit applications. The article analyzes the positioning accuracy grade distribution of RLS real-time high-precision positioning algorithm in detail. The analysis results show that the positioning accuracy is mainly affected by the error of frequency measurement, position error and velocity error, and the error of frequency measurement has a great influence. When the satellite position error is 5m,the velocity error is 0.01m/s, and the Doppler error is 0.1Hz, the probability of positioning error less than 1000m is 95.5%, which satisfies the normal distribution standard deviation 2σ distribution (95.5%).

With the development of wireless technology, Wi-Fi devices are extensively deployed in indoor environments. This fosters the development of Wi-Fi signal-based services and applications, e.g., indoor intrusion detection, human gesture recognition, indoor localization. However, the indoor environments are often complex and variable, Wi-Fi signals from transmitters through multiple paths to reach receivers. There is a large number of Non-Line-Of-Sight (NLOS) paths between the transmitter and the receiver, which causes seriously signal fading, deteriorating the quality of communication links, decreasing the accuracy of recognition application, and increasing the complexity of systems. In this study, an NLOS identification based on the wavelet packet transform (NIWPT) method is proposed. First, NIWPT collects raw channel state information (CSI) signals on the physical layer in current links. Then, NIWPT applies threelayer wavelet packet decomposition on the amplitude of CSI. A set of the wavelet packet coefficient, wavelet packet energy spectrum, information entropy, and logarithmic energy entropy as a feature vector is acquired. After that, the support vector machine is utilized to identify NLOS paths in the current links. Compared with other methods, NIWPT does not need to pre-process the raw CSI signals, it not only maximally reserves influence of the environment on the propagation signal, but also reflects the indoor environment more truly. The experimental results indicate that the recognition accuracy rate of the NIWPT method is 96.23% and 94.75% in the dynamic and static environments, respectively. It proves that the proposed method can effectively identify NLOS paths and has high identification accuracy and universality.

We solve the problem of video object segmentation by investigating how to expand the role of convolution in convolutional neural networks. Based on the One-Shot Video Object Segmentation (OSVOS) which can successfully tackle the task of semi-supervised video object segmentation, we introduce U-shape architecture. We first build a Global Guidance Module (GGM) on the bottom-up path to provide location information of potentially significant objects for layers of different feature levels. Then we design a Multi-scale Convolution Module (MCM) to fully get feature information and a Feature Fusion Module (FFM) to make the coarse-level semantic information well fused with the finelevel features from the top-down pathway. GGM and FFM allow the high-level semantic features to be progressively refined, yielding detail enriched segmentation maps. The experimental results on DAVIS 2016 data set shows that our proposed approach can more accurately locate the segmentation objects with sharpened details and our model has improved on all indicators than OSVOS.

Aiming at the problems of high cost and complex deployment of traditional human behavior recognition method system, a method for obtaining channel state information (CSI) for human behavior recognition using commercial Wi-Fi equipment is proposed. Using the amplitude and phase characteristics in the CSI as the base signal, the power spectrum entropy is used as a new feature to build a fingerprint library. The support vector machine (SVM) based on artificial fish swarm algorithm (AFSA) is used to classify and identify the action. The optimization of the classification is achieved by optimizing the parameter penalty factor and kernel function parameters in the SVM. According to the verification of real environmental data, the average recognition rate reached 94.64%.

For coherent integration detection of ultrafast maneuvering targets with modern radar, a novel long-time coherent integration algorithm, Polynomial Rotation-Polynomial Fourier Transform (PRPFT), is proposed to compensate across range unit range walk (RW) and Doppler frequency migration (DFM) simultaneously caused by super-high speed and strong maneuvering. First, RW can be corrected by the polynomial rotation transform (PRT) via rotating the coordinate locations of echo data. Then, the polynomial Fourier transform (PFT) can realize the compensation of DFM and coherent integration. To reduce the computational complexity, one decision method is proposed to search the multidimensional parameter space. Finally, numerical experiments are provided to validate the effectiveness of the proposed method.

Watermarking is of vital importance for copyright protection and content authentication of images. With the development of compressive sensing, it has been successfully applied for watermarking with improved performance. Since an image can exhibit tree structure in wavelet domain, a new watermarking embedding and extraction method is proposed based on tree-structured Bayesian compressive sensing. The Markov Chain Monte Carlo (MCMC) method and the variational Bayesian (VB) analysis can be used for inference, respectively. Attacks to the watermarking, such as Gaussian noise, salt and pepper noise, Gaussian filtering, and JPEG compression, are given to evaluate the watermarking robustness with comparison to other reported reconstruction algorithms such as basis pursuit, orthogonal matching pursuit, Bayesian compressive sensing using relevance vector machine (RVM), and Bayesian compressive sensing with VB. Simulation results and comparisons show remarkable advantages of the tree-structured Bayesian compressive sensing for watermarking embedding and extraction.

Terahertz (THz) technology is increasingly being used in a wide range of applications, and terahertz radar systems have also been developed in radar applications. In this paper, the terahertz radar system is used for 2 dimensional (2D) realtime imaging in near-field scenario within 20m. A real-time imaging system of 170GHz Synthetic Aperture Radar (SAR) is designed, and the system is simulated and verified by Doppler Beam Sharpening (DBS) algorithm. The simulation results show that the system can utilize uniform linear motion to synthesize a short aperture in the near-field range and form 2D image of scattering points in the scene. The imaging effect is good.

Proposed is a novel affine projection algorithm (APA) based on the M-estimate objective function with L₀ norm constraint. APA degenerates severely in impulsive interference and has no advantage for sparse system identification. In this letter, we use an M-estimate objective function to improve the robustness of the APA against impulsive interference, and a L₀ norm cost to improve the convergence rate for a sparse system. Simulation results show that the proposed algorithm outperform traditional algorithms in sparse system identification experiments that include correlated input and impulsive interference.

The blind image quality assessment algorithms produced every year are mostly “opinion-aware” (OA). It means that they require large numbers of subjective quality scores for regression model training. Subjective quality scores are not easily available, so people are eager to design an opinion-unaware (OU) algorithm which has free subjective quality scores. Besides, the OU algorithm has greater generalization capability than the OA algorithm. Therefore, we propose an OU algorithm based on a visual codebook for multiply distorted image quality assessment. Extensive experiments conducted on the three databases demonstrate that the proposed method is superior to the existing five OU methods in terms of the coherence with the human subjective rating. The MATLAB code is available at https://tonglewang.github.io.

Aiming at the requirement of the range and azimuth resolution of the near range target detection system, this paper completed the design of wideband real-time SAR imaging system. The system adopts “FPGA+RFADC+RFDAC” architecture, it can timely generate LFM(Linear frequency modulation) signal in L/S frequency band , complete the preprocessing of radar echo signals and realize the real-time SAR imaging algorithm, finally upload the imaging results in real time. The effectiveness of each function of the system is verified by simulation and actual measurement.

Plenty of natural language processing tasks can be modeled as text matching, such as question answering, machine translation and so on. An elementary and efficacious method is to distill matching patterns from words, phrases and sentences to obtain the matching score. In this paper, Reconstructed Color Interaction Image (RCII) is proposed to convert text matching to color image recognition. First, two texts are reconstructed and similarity operations are adopted to generate Color Interaction Image (CII). Then CNN is applied to extract hierarchical and elaborate matching information. Finally, fully connected layers are employed to obtain the matching score. Experiments have proved the effectiveness of our method.

In this study, we propose two feature extraction methods based on Electrocardiogram (ECG) RR intervals for diabetes Mellitus (DM) detection, respectively on the time and space dimension. Method Ⅰ is based on the pRRx sequence to detect diabetes subjects via signal recordings, which yielded the highest prediction precision value of 86%. Method Ⅱ is a new method of meshing Poincaré plot to extract the whole information entropy 𝐻𝐻(𝑋𝑋) and region information entropy 𝐻𝐻(𝑋𝑋)′ on the space dimension as features. When the grid gap of the meshing Poincaré plot is set as 50 and 400, we got the highest prediction precision value of 96%, which have better effect on the perspective of prediction accuracy comparing with method Ⅰ. In the future, we will collect more data of diabetic patients with our new improved ECG monitor to further optimize and improve the above feature extraction methods.

For the high-speed moving target, its high-resolution range profile (HRRP) obtained by wideband radar is stretched by the high order phase error. To obtain well-focused HRRP, the phase error induced by target velocity should be compensated, utilizing either measured or estimated target velocity. When the radar echo is under sampled, however, the HRRP will suffer from strong side and grid lobes, which deteriorates the performance of velocity estimation. A novel velocity estimation and compensation of high-speed target for under sampled data is proposed. The variational Bayesian inference based on the Laplacian scale mixture (LSM) prior is utilized to reconstruct HRRP with high resolution from the under sampled data. During the reconstruction of HRRP, the minimum entropy-based Newton method is used to estimate the velocity to compensate the high order phase error. Experimental results validate the effectiveness of the proposed velocity estimation and compensation algorithm.

In order to solve the problem that local binary pattern (LBP) is easy to lose some details when extracting facial features and image rotation leads to low recognition rate, a most value averaging LBP combined with gray level co-occurrence matrix feature algorithm is proposed. The method uses the most value averaging LBP algorithm to extract image features and reduces the feature dimension by principal component analysis (PCA); at the same time, considering the gray level co-occurrence matrix feature of the image, the most value averaging LBP feature is combined with the gray level cooccurrence matrix feature, and the k-nearest neighbor method (KNN) is used to classify and identify the face in lowdimensional space. The experimental results show that the proposed method has a good recognition effect.

Here we present a non-dispersive infrared (NDIR) methane sensor with semi-elliptical gas cell for monitoring the lower explosion limit (LEL) of methane (5%VOL in China). The special design of gas cell is used to reduce optical loss. And the novel NDIR scheme with narrowband mid-infrared light emitting diode (mid-IR LED) efficiently can solve the problem of humidity and background gas interference in theory and practice. This can be attributed to that the combination of narrowband LED and broadband photodiode (PD) was chosen. And insensitivity to humidity has been validated by experiments. Temperature dependency has been improved by temperature compensation. This scheme is not only used to detect the methane concentration, but also to detect other gas such as carbon dioxide contents in the air if the LED is changed.

Relative position detection sensor of high speed maglev train is one of the most important sensors in train positioning and speed measurement system. There is a complex circuit structure inside the sensor. How to ensure the reliability of sensors is the key problem to ensure the safe operation of maglev train, it is necessary to detect and diagnose the faults of the sensor which has been replaced or just left the factory. Kernel principal component analysis (KPCA) is used to diagnose sensor faults in this paper. This method is based on sensor data. It has the advantages of simplicity, convenience and high accuracy. The simulation and experimental results show that this method has a good effect on sensor detection and diagnosis.

Detection of the 5-methylcytosine (5mC) is one of the requirements of modern epigenetics research. The main challenges in this issue are how to distinguish the 5mC site from the unmethylated ones in the chain of nucleotides, and determine its amount. To streamline the tedious operations in traditional bisulfite conversion (BC) and polymerase chain reaction (PCR) based detection methods, lots of graphene derivatives and electrochemistry (EC) sensors have been exploited. In this work, we would like to propose an electronic method for DNA methylation detection by using five testing single strand DNA (ssDNA) chains as proof-of-concepts, based on the glutaraldehyde modified liquid exfoliated graphene field effect transistor (LEG-FET). First of all, for the sake of identifying the 5mC site, the immunorecognition strategy is utilized and incorporated with LEG-FET working principle. That is, the methylation sites on testing ssDNA chains are first recognized by the fixed 5mC’s antibody (5mCab) molecules on the channel of LEG-FET, then they are transduced to the varied current between the electrodes of drain and source (I_DS). It is found, the changing ratios of I_DS (ΔI_DS/I^DS0) are in negative relation with the amount of 5mC sites (N_mC) at each of the ssDNA concentrations (C_ssDNA). When C_ssDNA is varied from 1 to 10⁶ pM, the slopes of the responding curves -ΔI_DS/I_DS0 vs. N_mC are increased from 0.54 to 3.70 %/N_mC; meanwhile, at each of constant N_mc, the slopes of -ΔI_DS/I_DS0 vs. C_ssDNA are also examined to proof the repeatability in DNA methylation detection.

In this paper, the mechanism of the open-gate AlGaN/GaN HEMT based sensors were discussed and the effect of the ratio of gate length (L_G) to source-drain distance (L_SD) on the transconductance (g_m) of the sensors was investigated. It was shown that the smaller L_G/L_SD of the devices would get a higher maximum g_m (g_m-max). However, when the gate voltage (V_G) increased to a certain extent, the g_m of the larger L_G/L_SD devices would be higher. The experimental results were demonstrated by further theoretical calculation and analysis which is beneficial to enhance the sensitivity of the AlGaN/GaN HEMT based chemical sensors and biosensors by improving the g_m of them.

AlGaN/GaN high electron mobility transistors (HEMTs) have more remarkable properties in application of microwave transistors for high power and high frequency. A less widely studied application is high sensitivity to detect a wide range concentration of glucose. In this work, a photo-electrochemically treated open-gate AlGaN/GaN HEMT biosensor for glucose detection was developed. Through photo-electrochemical treatment, a smooth and thin gallium oxide can be formed on the sensing region. The threshold voltage was changed from -3.3 V to -1.3 V at a swept gain voltage. And a maximum value of transconductance was obtained at the gate voltage of 0 V. Effective functionalization of 3- aminopropyltriethoxysilane (APTES) and immobilization of glucose oxidase (GOx) can be realized on the oxidized sensing region. The proposed sensor exhibited good current response to glucose concentration over a wide linear range with high sensitivity above 8.61 × 10⁵ μA/mM·cm². The performance of the fabricated biosensor demonstrates the possibility of using AlGaN/GaN HEMTs for high sensitivity glucose detection in biochemical application.

The improvement of the detection limit of gas sensors has always been the focus of sensor research. Compared with the improvement of hardware, the improvement of the algorithm is still relatively less. In this study, a dual-channel methane gas sensor system based on mid-infrared LED light source was designed. We apply the wavelet denoising algorithm to the high-frequency noise suppression of the sensor system, which achieves a 36dB signal-to-noise ratio improvement over the traditional low-pass filter, making the detection limit of the sensing system reach the level below 3ppm. We give an estimation method for the detection limit of the sensing system. The detection limits estimated by this theory are basically the same as those obtained by the Allen deviation analysis in the conventional method. Implementing better algorithms to improve sensor SNR in software can reduce the demands of improving sensor SNR solely from hardware improvements.

Current sensing plays an important role in electric energy measurement, relay protection, intelligent equipment control and other fields in smart grid. Compared with traditional current transformers, for example, Rogowski coils and Hall current sensors, Tunnel magnetoresistance (TMR) sensor has the advantages of input/output magnetic isolation, AC and DC current measurement ability, wide bandwidth, small size, et al. Accordingly, TMR sensors are applicable for various scenarios and meet multiple current measurement needs. Based on the analysis of the principle of TMR sensor, the structure of current sensor system based on tunnelling magnetoresistance is introduced. Then, the designs of filter circuit and temperature compensation circuit are given. Finally, the experimental results show that TMR current sensor has good performance for current measurement, indicating widely applications in the field of current measurement in the future.