Search

Target detection is an important part of synthetic aperture radar (SAR) image interpretation, and its performance directly affects the accuracy of localisation and target recognition results in the region of interest. However, the SAR system resolution largely determines the SAR image index and imaging quality, and the target detection performance is closely related to the SAR system resolution in addition to the detection method. To improve the detection performance, the correlation between system resolution and target detection is investigated in this study. Multiple SAR images are used to test a SAR system with reduced resolution obtained by multi-look processing. At the theoretical level, the effect of different resolutions on target detection performance is analysed. MSTAR images, containing image chips with centred targets of varying orientation, are used to analyse the accuracy of the results. By comparing the measured experimental data with the simulation results, it can be obtained that the detection performance is the best when the resolution is roughly equal to the target scale.

Synthetic aperture radar (SAR) target feature extraction is a key technology for SAR target recognition. The existing SAR feature extraction methods mainly focus on amplitude information of the SAR image in spite of phase information. This study proposes a new feature extraction method introducing from concept of wavefront in optics domain. The Zernike polynomials are utilised to approximate the SAR echo wavefront, and these coefficients of Zernike polynomials are used as a new target feature for aiding SAR target recognition. Moreover, an experiment is carried out to prove that this method is effective about a target consisted of two scatters with different spacing between them.

Automatic targets recognition (ATR) for synthetic aperture radar (SAR) image is very important. ATR can be used in traffic management, national frontier safety, and so on. Traditional algorithms for SAR ATR is composed of extraction features and training classifier. The features are essential for the classification accuracy. However, choosing good features by hand is a hard task. The deep convolutional neural networks (CNNs) which can learn features automatically have got a great performance in natural images. However, the CNNs have many parameters and need a lot of data to train such networks. The remote-sensing data of SAR is limited. Then, the authors need a simple network which needs not much data and easy to train. The principal component analysis network (PCANet) is a shallow network that performs well in the recognition task and needs no hand features choosing. Though this network has produced a wide application in the natural images, it is rarely used in the SAR images. The experimental result of the moving and stationary target acquisition and recognition (MSTAR) dataset shows that the PCANet can achieve over 99% accuracy on ten-class targets. This result is better than traditional algorithms and is very close to the results of deep-learning methods.

The computational cost of many deep convolutional neural networks (CNNs) for the automatic target recognition proposed in the area of synthetic aperture radar (SAR) imagery recently is huge, and the limited SAR images are always insufficient for training the deep CNNs. To improve the computational efficiency, a new light but very efficient convolutional network architecture is designed using some novel techniques to get the better results. The authors apply the batch normalisation before each convolutional layer in order to reduce ‘internal covariate shift’ and use the drop-out strategy in the fully layer to avoid over-fitting. Additionally, concatenated ReLU is used as activation scheme specially instead of ReLU for preserving the negative phase information to get the double feature maps of the previous layer rather than to increase the depth of the filters that can lessen the parameters of the networks. The results of the experimental demonstrate that the authors’ CNNs can both achieve a state-of-the-art classification accuracy of 99.53% of the SAR target classification in the Moving and Stationary Target Acquisition and Recognition ten classes public dataset and perform well even when the training data is sparse.

The authors report a target separation detection and motion parameter estimation method based on the time-varying autoregressive (TVAR) model. The TVAR model is used to extract the instantaneous Doppler frequencies of multiple targets or scattering points in a same radar range cell, and the motion parameters are estimated based on the Doppler frequencies at every pulse time in a radar frame. In particular, for coherent multipulse echo signal, the TVAR model is first utilised to extract multiple instantaneous Doppler frequencies at each pulse time, thus forming a Doppler frequency matrix. Second, the Doppler tracking and polynomial fitting methods are utilised to estimate the radial velocity and acceleration based on the Doppler frequency matrix. Finally, the target separation detection is achieved by acquiring and validating multiple Doppler frequency components in the same range cell. Simulations and verifications are carried out, and the results show that the proposed method is effective for target separation detection and precision motion parameter estimation, which could be of great value radar target detection, tracking, and automatic target recognition.

To reduce the computation burden of inverse synthetic aperture radar on rotating targets imaging, a fast distributed simultaneous multiple orthogonal matching pursuit method is proposed in this study. In the proposed method, the distributed matching sparse representation model of the rotating targets is firstly established. Also then the total number of iterations is reduced by using multi-atom recognition strategy. Simultaneously, the computational complexity of each iteration is also reduced through QR decomposition. High-resolution image of the rotating targets can thus be obtained efficiently. Theoretical analysis and simulation results verify the effectiveness of the proposed method.

In this study, ground target recognition based on one-dimensional convolutional neural network (CNN) is studied by exploiting the targets’ high-resolution range profiles (HRRPs). Contrary to conventional methods which need feature extraction artificially, CNN can automatically discover features for classification. The authors propose a multi-channel CNN architecture that can be applied on diverse forms of HRRP such as amplitude, complex, spectrum etc. Experimental results demonstrate the superiorities of the proposed method over conventional methods based on handcrafted features and single-channel CNN in terms of recognition accuracy. Visualisation of the ‘deep features’ shows higher separability than handcrafted features, thus providing an insight into its effectiveness in exploiting the intrinsic structures.

Micro-Doppler (m-D) effect is a unique signature of radar targets with micro-motion, which provides a new technique for target recognition. Passive radar based on communication signal has a great significance in detection and recognition for micro-motion targets, low-altitude surveillance and control, national air defences etc. In this study, based on orthogonal frequency division multiplexing communication signal, the echo modelling and analysis of rotating target are operated, and the expression of the target m-D is deduced. Furthermore, the rotation frequency is extracted by sinusoidal frequency modulation Fourier-Bessel series. The feasibility of the proposed method is proved by the simulation results. The method could offer a reference to feature extraction of micro-motion targets under passive radar.

Synthetic aperture radar (SAR) target recognition can provide effective target category information and becomes the key part of SAR image application. Machine learning and deep learning are two main methods of target recognition. Normally, through the massive image features learning, the trained model can be used to infer the possible categories for various new target images, but the overfitting problem caused by limited data samples always makes the trained model unusable. In order to solve this case, the authors introduce a dual-input Siamese convolution neural network to the small samples oriented SAR target recognition. The training method looks like a kind of data enhancement method, but there are some differences between them. In the experiments, only 15 training samples are used to complete a three-class tank classification task. It means each category has just five samples while the number of corresponding testing data is 195. As a result, the recognition accuracy of the authors’ method outperforms the support vector machine, A-ConvNet, and 18-layers ResNet by 31, 13 and 16%, respectively. Siamese network has a good performance in small samples classification and the results prove the validity of the network.

This study considers the problem of coarse classification of targets using multifunction radar. Several methods are available for classification such as decision trees, Dempster–Shafer, Bayes, neural networks, etc. A different approach to assign the mass functions based on fuzzy logic in the Dempster–Shafer framework is proposed in this study. The method is evaluated for classification of different kinds of targets like aircraft, ballistic missiles, satellites, chaff and actual clouds, and unknown targets. With the proposed method, improvement in classification accuracy is observed, compared to existing mass functions. The technique is found to be computationally efficient and suitable for real-time systems.

In this study, a new method based on regularised orthogonal matching pursuit (ROMP) decomposition is proposed to extract the feature of the ballistic target's rotating micro-motion. Based on the establishment of the ballistic target rotation model, the general model of target's narrowband radar echo is obtained. An over-complete dictionary of atoms is established by combining the intrinsic characteristics of the signal model, and the selection of the matched atoms is carried out according to the regularisation principle. The extraction of the target's rotating micro-motion feature and the reconstruction of the original signal are realised. Simulation results show that the method can extract the micro-motion feature efficiently and accurately.

The combination of polarisation and high-resolution technology is a promising research direction for radar automatic target recognition. Fusing the polarisation information into the scattering centre model is able to refine the scattering structural information. This study proposes a fully-polarised radar high range resolution profile (HRRP) scattering centre estimation method based on a multiple measurement vector atomic norm minimisation model. The estimation is performed in a continuous domain, thus avoids the basis mismatch problem. The experimental results demonstrate the superiorities of the proposed method in terms of accuracy and robustness.

High-resolution range profile (HRRP) has received intensive attention in the radar automatic target recognition filed. Here, deep forest is applied to the recognition of HRRP. The deep forest is a deep learning method, which is a cascade of ensemble learners. In each layer, there are various ensemble learners. The input of each layer is the combination of the previous layer output and the original input data. The number of cascade levels can be adaptively determined such that the model complexity can be automatically set. Experiments based on measured data show that deep forest is feasible and achieves highly competitive performance for HRRP recognition.

Most existing algorithms in high-resolution range profile recognition focus on the closed set cases, where the test sample is from a known class. However, a sample could be drawn from unknown classes in realistic scenario, which is named as open set recognition. Here, open set HRRP recognition is achieved by incorporating extreme value theory into convolutional neural network. The softmax layer is replaced by a so-called openmax layer which estimates probabilities of the test sample belonging to known and unknown classes. Experimental results demonstrate that the proposed method outperforms the state-of-art algorithms such as NN, 1-vs-set machine, and W-SVM in terms of correct rejection rate.

Aiming at the application requirements of three-dimensional (3-D) fusion imaging of network radar, this study proposes an ISAR imaging resource-scheduling algorithm for network radar. Based on the feature recognition of each radar to each target, a projection-based method is first used to fuse the 3-D information of targets. Then according to the fused information, the demand of time resources for each radar is analysed. The radars selected for each target imaging are determined. Finally, the scheduling model of the imaging task for network radar is established, and time resources for each radar are randomly allocated. The simulation results show that the scheduling success rate of the proposed algorithm is two times on average than that of the traditional algorithm under the situation when the resources are near-saturated for both algorithms, which can improve the efficiency of network radar.

It is generally considered that increasing the carrier frequency of radar is an important way to improve the precision of micro-motion measurement. However, the increase of the radar frequency may raise the phase noise intensity of the radar transmitting signal and make the extraction more difficult, therefore it is particularly necessary to study the influence of phase noise on the extraction of micro-motion characteristics. In this study, specific studying about the effect of phase noise on the extraction of micro-Doppler (m-D) features is carried out. The effect of phase noise on the extraction performance of the m-D features is evaluated based on a parameter micro-motion signal-to-clutter ratio and an empirical formula is put forward based on the experiments. The results of simulation experiments indicate that increasing the carrier frequency cannot improve the extraction performance of micro-motion features in the case of using both time–frequency analysis method and the new developed sinusoidal frequency modulation Fourier transform (SFMFT) method. Meanwhile, increasing the frequency of target's micro-motion has an improvement effect through time–frequency method due to the increase of Doppler frequency.

Polarimetric information is of great importance for radar target recognition. Conventional polarimetric features are hand-designed based on scattering mechanism. In this study, a novel polarimetric target recognition framework based on long–short-term memory (LSTM) network is proposed. The different polarimetric channels are regarded as the sequential inputs in LSTM, and the features are extracted automatically. Experimental results on dual-polarised high-resolution range profile recognition demonstrate that the features learnt by LSTM are more discriminating than conventional features. The recognition performance of the proposed method outperforms the state-of-the-art methods as well.

With the increase of radar signal bandwidth and the application of RF directing sampling technology, wideband radars are facing greater pressures for data transmission, storage, and processing. Based on the relationship between linear frequency modulation (LFM for short) signal and LFM-stepped signal, synthesis method of high-resolution range profile (HRRP for short) for LFM step signal can be applied to LFM signal. Combining the synthesis method of HRRP for LFM signal with compressed sensing theory, this study proposes a new compression method for wideband radar echo with LFM signal style based on random segment selection and analyses the constraint conditions of the proposed data compression method. Finally, the validity and practicability of the proposed data compression method are verified from two aspects: simulation experiment and actual data processing.

An end-to-end radar high-resolution range profile recognition method is proposed based on stacked denosing sparse autoencoder which stacks several denosing sparse autoencoders and uses softmax as the classifier. The training process consists of two steps. The first is layer-by-layer pre-training and the second is fine tuning using the pre-training results for initialisations. The two-step training process makes this model converge faster and more likely to converge to the global optimal point than directly training the joint network. Experimental result shows that the proposed method achieves higher recognition accuracy than state-of-art methods.

This study develops a robust method for ship target recognition, based on inverse synthetic aperture radar (ISAR) image matching with the template of 3D model. To deal with the difficulty of building template lib, the authors develop to use CAD model instead of the live radar data, which can be projected to the range–Doppler plane at any angle. In the study, the method of matching between the observed ISAR image and candidate target image of model is also proposed, including feature extraction and scoring for recognition. In addition, influences of estimation accuracy for projection are analysed and simulated. Finally, recognition results obtained with simulated data are provided to verify the effectiveness of the proposed approach.

Radar cross section (RCS) is an important characteristic of radar targets. The mean, variance, skewness, kurtosis, varying patterns of RCS sequences provide rich features for radar target classification. In this study, an RCS classification method based on Kullback–Leibler divergence (KL divergence) is proposed. KL divergence is a measurement of the distance between two probabilistic distributions p(x) and q(x). In this method, the discrete probability distributions of the training RCS sequences and their varying patterns are firstly calculated and saved. For a target with unknown label, the corresponding probability distributions of its RCS sequence are computed. Then, the KL divergence between this distribution and the pre-trained distributions is also computed. The label of this target is determined by these KL divergences and an adaptive threshold. The method collects statistical features and dynamic features of RCS sequences, and provides a unified solution for classifying RCS sequences, which is of great importance for radar target recognition. The efficiency of this method is verified by experiments on simulated data.

Jet engine modulation (JEM), a modulation phenomenon induced by the rotating structure of jet engines, is a representative feature extracted from the radar returns for aircraft target recognition. As one of the most significant periodic characteristics of JEM, the chopping frequency plays an important role in aircraft target recognition. However, existing methods either need complicated logical algorithms to get an accurate estimation of the chopping frequency or exploit the chopping frequency as priori information for further feature extraction, and neither of these two scenarios is practical. Here, the authors propose a robust and fast algorithm for extracting the chopping frequency, which uses wavelet decomposition combined with autocorrelation to process the analytic form of the JEM signal. Application results of measured JEM signals demonstrate that the proposed algorithm is effective and practical for stable and accurate extraction of the periodic feature from JEM signals.

The synthetic wideband signal has the advantages of high range and velocity resolution. It has a wide range of applications in target detection and recognition. While in the processing of signal generation, due to the improper setting of hardware system parameters, the bandwidth of the transmitting signal may not be consistent with the nominal value, which will affect the result of signal processing. In this study, the influence of system bandwidth error on synthetic high-resolution range profile is theoretically analysed. On this basis, an error compensation method based on parameter traversal is proposed and verified with simulation and experimental data processing.

Target recognition is a challenging problem in air-to-surface terminal guidance for missile-borne millimetre wave radar (MWR). The class of high range resolution profile (HRRP)-based automatic target recognition approaches is sensitive to the attitude of target. One effective way to improve the performance of target recognition is acquisition of multidimensional information from target's image. Here, 2D ISAR imaging of the ground static target for the forward-looking missile-borne MWR is studied. The relative signal model and some key points are analysed. The effectiveness of the proposed approach is demonstrated by simulation and real data processing results.

An effective Max-Fire CNN model MF-SarNet for synthetic aperture radar (SAR) automatic target recognition (ATR) is presented, here. By selecting the convolution kernel of the Fire module in the network, the parameters are reduced to obtain the effective convolutional neural network of less parameter. In view of the requirement of deep learning for large-scale data, an augmentation method is proposed, which can learn the features of large database better. The results based on MSTAR database show that the model is effective and the result is encouraging. The accuracy of SAR image recognition is 98.53%.

The radar high-resolution range profile (HRRP) is of great significance for target recognition in the ground compound environment. As two essential processes of HRRP target recognition, discrimination and classification performance improvement are studied in this study. A novel target recognition method is designed, denoted by multiple support vector (SV) method. In the training phase, a multiple correlate SV model is constructed, which separates the multi-target feature space into several central-coherence distributed regions. Each region has been described by the modified SV method, and the category determination is based on the description of multiple regions. This method can overcome the issues of both outlier discrimination and targets classification. Due to the hierarchical process in the testing phase, this method needs less memory and calculation to fit the practical radar target recognition task. Finally, the experiment based on the measured data verifies the excellent performance of this method.

Traditional synthetic aperture (SAR) imaging methods assume that man-made targets are composed of point-based scattering centres. In this way, gaps of components may be resulted, because of neglecting their inherent scattering behaviours. In order to enhance the integrity of components in SAR, a feature-enhanced (FE) imaging method for SAR is proposed, taking into account the inherent behaviours of components of radar targets. Here, the radar echo is decomposed into attributed scattering centres (ASCs) instead of traditional point scattering centres. Based on the decomposed ASCs, different feature enhancement schemes are applied to different kinds of scattering centres according to their attributed parameters. As a result, a FE imaging image for SAR can be achieved by utilising the inherent feature of each decomposed ASCs.

Utilising neural networks to learn and extract valuable features has achieved satisfactory performance for synthetic aperture radar automatic target recognition (SAR ATR). However, such target recognition capability could be seriously limited by severe image rotation. To greatly improve the performance of convolutional neural networks-based SAR ATR, a data augmentation method combining region of interest (ROI) extraction and full-angle rotation method is firstly proposed in this study. Then, an inception-SAR NET is presented to polymerise multi-branch feature maps. The superior performance of inception-SAR NET structures is obtained by comprehensive experiments. Finally, the results based on MSTAR dataset demonstrate that authors’ methods could achieve the most advanced performance.

Radar moving target detection (MTD) technology is a key technology in the field of radar signal processing. The MTD method of radar has a better performance when applied to uniform motion detection although it has limited performance in other aspects, and it is also difficult to distinguish the type of the moving target. This article presents a new method for detecting and classifying moving targets based on convolutional neural network, which uses the convolutional neural network to learn the motion characteristics of the moving target echo so that realises the detection and classification of moving targets. At first, the authors model the moving target echo. Then the short-time Fourier transform (STFT) is used to perform time–frequency analysis. The obtained time–frequency graph is used as the input of the convolutional neural network which can automatically learns moving target features through training. Finally, the authors use the trained convolutional neural network model to detect and classify moving targets. The simulation verifies that the method has a great improvement in the detection accuracy rate compared with the traditional MTD.

With the development of synthetic aperture radar (SAR), it is more and more important to recognise the image obtained by SAR. SAR is a useful tool for geoscience and remote sensing, and there are a large amount of SAR images and products. How to quickly identify and process this massive information in the shortest time and find the target in time is the focus of research in this field. In the traditional processing method, the digital processing is used. It is time and storage consumption due to its massive data. This study proposes optical methods for target recognisation of SAR images. In this method, it takes the advantages of high speed and high resolution in optical processing methods to solve the bottleneck encountered by traditional methods. The method is based on an optical correlator. The optical processing method has a large capacity for a single processing and has a great advantage over the simple calculation method in processing speed. The experimental results demonstrate the effectiveness of the optical processing method.