The development of radar high resolution range profile(HRRP)non-cooperative targets recognition technology is mainly limited by two aspects:(1) Due to the low observation frequency of non-cooperative targets,the number of labeled HRRPs is insufficient,making non-cooperative HRRP based target recognition a typical few-shot recognition problem,which is still a hot and difficult issue without definite conclusion in the academia.(2) The existing HRRP based target recognition methods are mostly based on the hypothesis of complete dataset,making them mismatch with non-cooperative target recognition in few-shot setting.In this paper,we put aside the complete hypothesis and propose an HRRP based few-shot target recognition method with CNN-SSD.The proposed method first uses a complete training HRRP containing 45 classes of cooperative targets to learn an initial category-independent feature extractor,on the basis of which we further utilize the model sequential self-distillation mechanism to obtain a more generalized feature extractor.Finally,the generalization ability of the extracted features is evaluated on unseen non-cooperative targets during training.Experimental results on self-simulated HRRP dataset reveal that the proposed method can achieve an average recognition rates of 61.26%,84.69% and 92.52% respectively when only 1,5 and 10 annotated HRRPs of non-cooperative targets are available.
This paper focuses on the problem of radar targets detection in the compound-Gaussian sea clutter on the condition with the limited secondary data.The texture is modeled by the generalized inverse Gaussian distribution.Two adaptive detectors based on a priori knowledge of the speckle covariance matrix are proposed.First,the inverse complex Wishart distribution is exploited to model the speckle covariance matrix,and then an adaptive detector without using the secondary data is designed according to the generalized likelihood ratio.According to the maximum posterior test criterion,the secondary data are used to design an adaptive detector with secondary data and prior knowledge.Experimental results show that when the number of secondary cells is small,the two detectors proposed in this paper have a better detection performance than the GLRT-GIG detector.With different numbers of secondary cells,the proposed adaptive detector depending on the secondary data and a prior knowledge has the best performance.
Optimal data association is the main task of multi-target tracking due to the similarity of the tracker’s filtering parts.Traditional Multi-target tracking methods pick up the optimal data association from all possible associations that account for the complexity exponentially increasing with the number of targets and limiting the maximum number of targets which can be stably tracked.This paper proposes an efficient and accurate method where the measurement points raised by targets and clutter are modeled as the Poisson point process and the expectation maximisation algorithm is utilized to estimate the target states recursively.Independent data association and mixing probability decrease the computational complexity.Furthermore,Doppler information refers to the fact that the target feature has been used in association and filtering stage to improve tracking performance without adding complexity.The experiment with simulation data show that the performance of the proposed method is better than that of the traditional method with a shorter operation time.
The ionospheric clutter is a time-varying,non-stationary and non-Gaussian complex clutter in the high-frequency surface-wave radar(HFSWR) system and its suppression is a daunting task.Due to the complexity of the ionospheric clutter,the single clutter suppression algorithm cannot suppress the clutter effectively.Classification of the complex ionospheric clutter according to different characteristics can improve the performance of ionospheric clutter suppression significantly.After a complete analysis of the characteristics of ionospheric clutter,this paper proposes an improved ionospheric clutter classification method based on fuzzy C-means clustering.Experimental results show that the proposed method has a better performance than the traditional algorithm in clustering the ionospheric clutter.
Recognizing the operation modes of airborne fire control radar quickly and accurately,especially in the high threat mod,is the important part of airborne electronic countermeasures.For the restricted application conditions of existing methods,a new method using amplitude rearrangement is proposed in this paper after a large amount of analysis of each mode.This method makes full use of the characteristics of amplitude distribution to identify tracking signals.First,all the signals are rearranged by the size of signal’s amplitude,and then the first-order difference of the rearranged amplitude iscalculated.Finally,the track and search mode and single target tracking mode are recognized effectively according to the special notch character of the rearranged amplitude.Simulation results prove the good real-time and accuracy rate performance of the proposed method.More importantly,the new method is not sensitive to the pulse deletion and interference.Owing to the above advantages,the proposed method has a good potential in the practical engineering application.
The traditional HRRP recognition methods do not consider the temporal correlation,and the azimuth sensitivity of HRRP results in the temporal variation of the samples.This paper proposes a multiplicative recurrent neural network.In this paper,HRRP samples are converted into the sequence form first,which is used to consider the correlation between range cells.In order to alleviate the mismatch between the HRRP sequence variation caused by azimuth sensitivity and the model with fixed parameters,the model adaptively selects the corresponding parameters according to the input data,and extracts robust features from the HRRP sequence.Finally,the voting strategy is adopted to fuse the information at all time steps and predict the sample categories.Experimental results with measured data show that the current model can effectively extract discriminative features and identify targets.
The returns of slow and dim targets,which have high energy and a wide frequency spectrum,are often submerged in the sea clutter.This paper proposes a new SVD-FRFT-based sea clutter suppression method using time-frequency information.The proposed method adopts the strategy of block processing and overall judgment.First,the long echo sequence is divided into sub-pulse blocks,with the first Q-order signal components of each sub-pulse block being respectively extracted and inter-block signal correlation performed to realize the preliminary separation of the target and clutter.Then an adaptive segmented SVD-FRFT processing is used to achieve further purification of the signal components of each order.Finally,the normalized ime-frequency ridge quadratic fitting error is used to judge the components of each order.In this way,the target is retained while the sea clutter is suppressed.The effectiveness and robustness of this approach is verified by the real sea clutter data.Compared with the classical sea clutter suppression methods,the proposed method can effectively suppress the sea clutter and extract target signals regardless of signal-to-clutter ratio(SCR),especially for the multi-target.
To address the problem that the clutter suppression performance is reduced for the multichannel synthetic aperture radar(SAR) because of the ambiguous component of the land clutter in the inshore area,we propose an inshore ambiguity clutter suppression method aided by on clutter classification.First,a multi-look interference feature covariance matrix(MIFCM) for each range-Doppler unit is constructed,which includes multi-look interference magnitude/phase and multi-look interference magnitude gradient information.Next,with the affine invariant riemannian(AIR) distance measure of MIFCMs,the clutter classification result can be acquired.Finally,with the clutter classification result,we select independent and identically distributed(IID) training samples to estimate the clutter covariance matrix in the ambiguous clutter region,with clutter suppression performed in the multichannel SAR image domain.Simulation result shows that the proposed method can accurately classify and suppress the azimuth ambiguous clutter in the inshore area.
The direct SAR imaging method has a low efficiency when used for realizing the target echo simulation and SAR imaging,which will meet,with difficulty,the requirement of rapid SAR generation in the closed-loop verification process of sea detection guidance.In this paper,we propose a SAR simulation method based on three-dimensional(3D) scatter centers.First of all,the 3D scattering centers of a ship target from different perspectives are obtained by the 3D fast imaging based on the ray tube integral and 3D scattering center extraction based on the CLEAN algorithm.Then,the target SAR echo data and image aggregation processing are quickly calculated to obtain the SAR complex image through the geometric modeling of SAR imaging from the corresponding perspective.The method makes comprehensive use of electromagnetic scattering center extraction and complete SAR processing flow,which can significantly improve the efficiency and flexibility of target imaging simulation under the condition of maintaining the electromagnetic calculation accuracy.The calculation accuracy,efficiency,and flexibility of the proposed method are verified by taking a typical ship target as an example.
Radial size estimation using radar high-resolution range profiles(HRRPs) and heading angle estimation are the main means for ship classification.The classification ability is closely related to the range resolution of the radar,precision of radial size estimation,and prior distribution of ship lengths in different offshore areas.We collected the AIS information on about 30 000 ships and their lengths in the four offshore areas of China in the ship information net of China.By fitting the data of ship lengths in each offshore area,it is found that the Weibull distributions provide good-of-fitness to the ship lengths and the parameters in individual area are rather different.Based on the prior distributions of ship lengths,we derived the quantitative relationship between the correct classification probability of big-moderate-small ships and the estimate error of ship radial size.The results indicate that the condition for the big-moderate-small correct classification probability in the offshore areas of China to be up to 90% is that the estimate errors of the ship radial size estimates falls into the interval(-12.67 m,9.41 m) when the heading angle of the ship is between ±75 degrees.
Aiming at the problem of strong ambiguity and uncertainty in the observed data,the author proposes a fuzzy data association algorithm based on the historical features of high-resolution one-dimensional range profile.First,for the high-resolution one-dimensional range profile's attitude,amplitude,and time-shift sensitivity,feature extraction is performed to obtain low-sensitivity features.Then,the features of the track initiation are used to construct the initial feature sample database;the features of the historical moment are utilized to construct the historical feature sample database,and the feature sample database is updated in real time.The feature weight is obtained by the interval entropy weight method and the fuzzy membership of the measurement,and the target is calculated to construct a fuzzy matrix.Finally,fuzzy data association is completed based on the principle of maximum fuzzy membership.Experimental results show that,in both the maneuvering and non-maneuvering scenarios of the target,the association performance of the proposed algorithm is better than that of the fuzzy data association algorithm.And with the increase in the clutter density,the association performance of the two algorithms is gradually decreased,but the association performance of the proposed algorithm becomes better.
Aiming at the spotlight mode of the distributed small satellite multiple-input multiple-output synthetic aperture radar(MIMO-SAR),an ultra-high resolution imaging method is proposed to reduce the storage pressure and imaging load of the satellite in a multi-channel high-resolution mode.The full-aperture signal of each channel is divided into sub-aperture signals.The sub-aperture signal is deblurred,and then the large bandwidth signal is obtained by using the improved time domain bandwidth synthesis method.Then the sub-aperture image coherent fusion algorithm is used to obtain the ultra-high resolution image.Simulation results show that the improved time domain bandwidth synthesis method can synthesize bandwidth effectively,and the imaging effect of the proposed method is excellent.The validity and reliability of the method are verified.
In the wide-angle synthetic aperture radar(SAR),the scattering behavior of many illuminated objects might vary with the observation angle,which results in the degradation of the resolution and interpretability of the reconstructed imagery.To solve this problem,a sparse-based source separation and imaging method is proposed in this paper.The distinct scattering behaviors of the isotropic and anisotropic scattering targets are employed to formulate a composite projection operator.Then,the sparse constraint is utilized to suppress the cross-projection energy and imaging of the mix-received wide-angle SAR data is realized.Finally,the imagery of the anisotropic scattering targets can be derived with improved focal quality and interpretability.Numerical simulation verifies the validity of the proposed methodology.The resolution and resolution of the reconstructed image have been significantly improved.
In the traditional scattering center extraction method based on the shooting and bouncing ray(SBR) technique,only the physical optics contribution of the target is considered.However,the physical optics method is unable to describe the contribution of the actual edge diffraction.By comprehensively considering the specular reflection and edge diffraction contribution of the target,an edge diffraction correction method for 3D scattering center modeling based on the SBR technique is proposed.Using the SBR technique and combining the image domain ray tube integration and equivalent edge currents method,the edge diffraction correcting formula for the 3D inverse synthetic aperture radar image is derived.Simulation results show that the proposed method is able to effectively improve the modeling accuracy of the 3D scattering center.
Aiming at the conflict game between multistatic multiple-input multiple-output(MIMO) radars and multiple targets,a joint technology of power allocation and beamforming for a multistatic MIMO radar system is studied.The main purpose of the multistatic radar system is to minimize the total transmit power under a certain signal to interference noise ratio(SINR) constraint,and to suppress the cross-channel interference to improve the accuracy of target detection.Based on the strategic non-cooperative game between radars,the existence and uniqueness of the Nash equilibrium solution are proved.Then,according to the theoretical analysis,a joint beamforming and power allocation algorithm is proposed and converges to the Nash equilibrium.Meanwhile,the receive beamformers of multistatic radars are obtained by using the linearly constrained minimum variance(LCMV) criterion to suppress the mutual radar interference.Numerical results show that the proposed algorithm has a better interference suppression ability and less power consumption than the related game algorithm.
For the existing jamming discrimination methods for multistation radar systems,only the single feature of target echo space correlation is utilized as the metric,which leads to insufficient comprehensiveness of feature extraction,so that effectiveness and universality are insufficient for the discrimination algorithm.In this paper,an identification method in multistatic radar systems based on the deep neural network is proposed.This method combines the characteristics of multistatic radar systems cooperative detection technology,which has many available resources and strong scheduling ability in space,time and frequency domain,with the strong model learning and feature representation ability in the process of information processing on the deep neural network,so that it can effectively apply to the field of anti-deception jamming.Full use is made of unknown information about echo data to obtain more multi-dimensional,more comprehensive,more complete and deeper feature differences besides correlation,so as to achieve a better jamming discrimination effect.Simulation results show that the proposed method can effectively reduce the influence of noise and pulse number on the jamming discrimination performance.At the same time,the limitation of the target echo correlation coefficient on anti-jamming technology under nonideal conditions is alleviated,which broadens the boundary conditions of the application process.
Interrupted sampling repeater jamming is a new type of coherent jamming that generates a single or multiple false target groups,which cannot be effectively suppressed using conventional signal processing methods.To solve this problem,the received signals of several pulse repetition cycles are first accumulated in the time-frequency domain to enhance the time-frequency information on the signals.Then,the distribution of jamming signal is extracted accurately on the time-frequency surface based on the image processing algorithm,and the target echo is roughly reconstructed according to the time-frequency domain filtering method.Finally,based on the reconstructed time-domain signal,a band-pass filter is constructed to suppress the interference and noise.Simulation results indicate that compared with other methods,this scheme can effectively suppress various types of interrupted sampling repeater jamming in a low signal-to-noise ratio and multi-target environment with robustness and accuracy.
In order to solve the problem of optimizing deployment for multistatic radar,an optimum deployment method adopting unequal barrier coverage is proposed.This method employs the relationship among the unequal coverage barriers to construct constraints,and divides the optimization problem into two sub-problems:the optimization problem of a single deployment line and the optimization problem of the combination of unequal barriers.Then,based on the optimal deployment condition of a single deployment line,the exhaustive method is used to solve this optimal deployment problem of a single deployment line.Moreover,according to the minimum deployment cost criterion,the integer linear programming is used to solve the optimization problem of the unequal barrier combination.Meanwhile the optimal deployment parameters are determined.Simulation results reveal that,compared with the existing similar methods,the proposed optimum deployment method can achieve lower deployment costs and requires fewer transmitters.
The reduction of the signal bandwidth and sampling frequency is conducive to the realization of antenna array thinness based on the micro system so as to improve the adaptability of the large aperture antenna on the ultra-high speed platform.However,the high-resolution range profile of the ground moving target is required for accurate tracking and recognition of the ground moving target.In order to solve this problem,a system architecture and processing method based on the low instantaneous bandwidth digital array of transceiver reciprocity is proposed,which can realize high resolution ground moving target display(GMTI) by frequency band synthesis.Because the array transceiver reciprocity can realize the coincidence of the equivalent phase center,the phase difference of the target deviated from the beam center does not need to be compensated in the process of bandwidth synthesis,and the precise synthesis of large bandwidth signals can be realized.After the synthesis of the wideband signal,the clutter is suppressed by space-time adaptive processing(STAP) to realize the detection of ground moving targets.Simulation results verify the correctness of the system architecture and processing method.
Based on the wide beam antenna radar level meter,an improved algorithm for wide beam linear array radar level meters is developed to solve the problem that the detection performance of wide beam antenna radar is extremely vulnerable to the strong background clutter.For a uniform linear array frequency modulated continuous wave radar,the first step of the algorithm consists of the detection of strong scattering points through range and estimation direction of arrival for strong scattering points;the second step is the recognition of efficient points from strong scattering points set on the basis of analyzing the production condition of strong scattering points.The measurement value was obtained from the efficient points set.A set of field experiments was conducted in a grain processing factory with a 77 GHz wide beam linear array radar level meter.The proposed algorithm can achieve an accurate and steady measurement of the low-dielectric constant material level in harsh industrial conditions.This study does away with the restriction of beam width,which is of great significance for the popularization and application of miniaturized,low-cost wide beam radar level meters.
Under the strong interference of respiration and its harmonics,a weighted reconstruction method of the second harmonic signal of the heartbeat is proposed to significantly improve the measurement accuracy of the heart rate by using the 77GHz FMCW radar.Based on the results of multi-resolution analysis of the chest wall displacement signal,the energy distribution characteristics of each layer is used to weight the sub-band within the second harmonic range of the heartbeat,which can highlight the most likely frequency range of the second harmonic of the heartbeat.Therefore,both the strong interference from respiration and its harmonics and the difficult problem of model order selection in the spectral peak estimation can be avoided.Experimental results from real measured data validate that the proposed method can achieve high-accuracy heart rate monitoring based on the 77GHz FMCW radar.
To address the problem of IQ imbalance in the demodulation of radar intermediate frequency(IF) signals,this paper proposes a method for digital IQ calibration in the frequency domain based on conjugate symmetry.By modeling the IF signal at the presence of IQ imbalance,its spectral characteristic is first analyzed.The IF signal in the frequency domain can be decomposed into the unilaterally ideal signal and bilateral error signal.Among these,the bilateral error signal can be further decomposed into two components,the mirror-frequency error and the main-frequency error.Then,the relationship between the original spectrum of these two components and the spectrum of their conjugate counterparts is used to conduct some simple calculations for obtaining a frequency-domain estimate of the error signal.Therefore,the estimated error can be used to make the compensation in the IF signal.Experimental results from synthetic aperture radar imaging verify that the proposed method can effectively suppress both the mirror-frequency error and the main-frequency error from the IF signal and calibrate IQ imbalance to achieve the high performance of imaging.
In order to meet the requirement of higher transmission power for the millimeter wave radar or communication system,a Ka band power amplifier is designed based on the 65 nm bulk Si CMOS process.This power amplifier works at 30~32 GHz ,and consists of two-stage CASCODE differential pairs structure amplifiers.Neutralizing capacitances are used to enhance its stability,and the on chip matching network is realized based on transformer coils.After testing,the maximum output power of the power amplifier in the operating frequencies is 16.3 dBm.Its maximum PAE is 16.9 %,-1 dB compression point is 13.2 dBm,and power gain is 23.6 dB.The power amplifier chip presented in this paper has advantages in power gain and chip area utilization,which provides a feasible high power output design example for the silicon-based millimeter wave power amplifier.
Cross-camera target tracking is very challenging,mainly because of the difference in the background area under different cameras and the randomness of the target movement behavior trajectory,which will accumulate interference errors very easily,and affect the matching accuracy,thus leading to the tracking failure.Aiming at this problem,a model of moving target tracking based on sparse representation is proposed in this paper.The model uses the difference in background brightness between different cameras to compensate the illumination of the target,so as to obtain a stable template matrix.At the stage of model solution,to solve the problem that the traditional greedy algorithm has a single atom matching pattern,ignoring the relationship between inner atoms and leading to a low reconstruction accuracy,the model adopts the band exclusion(BE) method in the band exclusion local optimization orthogonal matching pursuit(BLOOMP) algorithm to reduce the interatomic coherence.In addition,combining the local optimization(LO) technique with the new coherence discrimination mechanism,we obtain a more compact correlation band to update the support set,leading to improving the reconstruction accuracy.At the stage of template updating,in order to enhance the real time performance of the template matrix,the model uses correlation band and different weight coefficients as the template replacement mechanism.Simulation results show that the proposed method can track the interested target stably and robustly compared with the traditional algorithm under the condition of indoor and outdoor scenes.
To improve the accuracy of cross-modal pedestrian re-identification,a reciprocal bi-directional generative adversarial network-based method is proposed.First,we build two generative adversarial networks to generate cross-modal heterogeneous images.Second,an associated loss is designed to pull close the distribution of features in latent space during the image translation between visible and infrared images so as to help the networks generate fake heterogeneous images that have high similarity with the real images.Finally,by concatenating the original and generated heterogeneous pedestrian images into the discriminative feature extraction network,images from different modalities can be unified into a common modality,thus suppressing the cross-modal gap.Representation learning and metric learning are utilized to achieve more discriminative pedestrian features.Comparative experiments are conducted on SYSU-MM01 and RegDB datasets to analyze the accuracy with different loss functions.Compared with other state-of-the-art cross-modal pedestrian re-identification methods,the proposed method achieves a higher accuracy and stronger robustness.