Zongjie Cao

University of Electronic Science and Technology of China, Hua-yang, Sichuan, China

Are you Zongjie Cao?

Claim your profile

Publications (24)10.36 Total impact

  • [Show abstract] [Hide abstract]
    ABSTRACT: Cognitive robotic systems nowadays are intensively involving learning algorithms to achieve highly adaptive and intelligent behaviors, including actuation, sensing, perception and adaptive control. Deep learning has emerged as an effective approach in image-based robotic perception and actions. Towards cognitive robotic perception based on deep learning, this paper focuses the Constrained Restricted Boltzmann Machine (RBM) on visual images for sparse feature representation. Inspired by sparse coding, the sparse constraints are performed on the hidden layer of RBM to obtain sparse and effective feature representation from perceived visual images. The RBM with Sparse Constraint (RBMSC) is proposed with a generalized optimization problem, where the constraints are applied on the probability density of hidden units directly to obtain more sparse representation. This paper presents three novel RBM variants, namely L 1-RBM, L 2-RBM, and L 1/2-RBM constrained by L 1-norm, L 2-norm, and L 1/2-norm on RBM, respectively. A Deep Belief Network with two hidden layers is built for comparison between each RBM variants. The experiments on MNIST database (Mixed National Institute of Standards and Technology database) show that the L 1/2-RBM can obtain more sparse representation than RBM, L 1-RBM, L 2-RBM, and Sparse-RBM (SRBM) in terms of sparseness metric. For further verification, the proposed methods are still tested on MNIST Variations dataset. The recognition results from perceived images in MNIST and MNIST Variations demonstrate that our proposed constrained RBM variants are feasible for object cognitive and perception, and the proposed L 1/2-RBM and L 1-RBM outperforms RBM and SRBM in terms of object recognition.
    Journal of Intelligent and Robotic Systems 02/2015; DOI:10.1007/s10846-015-0213-3 · 0.81 Impact Factor
  • Zongjie Cao · Yuchen Ge · Jilan Feng
    [Show abstract] [Hide abstract]
    ABSTRACT: Since the traditional CFAR algorithm is not suitable for high-resolution target detection of synthetic aperture radar (SAR) images, a new two-stage target detection method based on variance weighted information entropy is proposed in this paper. On the first stage, the regions of interest (ROIs) in SAR image is extracted based on the variance weighted information entropy (WIE), which has been proved to be a simple and effective quantitative description index for the complex degree of infrared image background. Considering that SAR images are nonuniform, an experiment is conducted ahead, in which the value of the variance WIE from a real SAR image in three areas with significant different uniform levels are tested and compared. The results preliminarily verified that the variance WIE is able to measure the complex degree of SAR images. After that, in order to make the segmentation efficient, the rough ROIs are further processed with a series of methods which adjust ROIs into regular pieces. On the second stage, for each of the ROIs, a variational segmentation algorithm based on the Split-Bregman algorithm is adopted to extract the target. In our experiment, the proposed method is tested on two kinds of SAR images, and its effectiveness is successfully demonstrated.
    Journal on Advances in Signal Processing 11/2014; 2014(1). DOI:10.1186/1687-6180-2014-45 · 0.81 Impact Factor
  • Jilan Feng · Zongjie Cao · Yiming Pi
    [Show abstract] [Hide abstract]
    ABSTRACT: In recent years, sparse representation-based techniques have shown great potential for pattern recognition problems. In this paper, the problem of polarimetric synthetic aperture radar (PolSAR) image classification is investigated using sparse representation-based classifiers (SRCs). We propose to take advantage of both polarimetric information and contextual information by combining sparsity-based classification methods with the concept of superpixels. Based on polarimetric feature vectors constructed by stacking a variety of polarimetric signatures and a superpixel map, two strategies are considered to perform polarimetric-contextual classification of PolSAR images. The first strategy starts by classifying the PolSAR image with pixel-wise SRC. Then, spatial regularization is imposed on the pixel-wise classification map by using majority voting within superpixels. In the second strategy, the PolSAR image is classified by taking superpixels as processing elements. The joint sparse representation-based classifier (JSRC) is employed to combine the polarimetric information contained in feature vectors and the contextual information provided by superpixels. Experimental results on real PolSAR datasets demonstrate the feasibility of the proposed approaches. It is proven that the classification performance is improved by using contextual information. A comparison with several other approaches also verifies the effectiveness of the proposed approach.
    Remote Sensing 07/2014; 6(8):7158-7181. DOI:10.3390/rs6087158 · 3.18 Impact Factor
  • Jilan Feng · Zongjie Cao · Yiming Pi
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents an SAR image classification approach that takes advantage of both amplitude and texture features. The proposed approach is based on superpixels obtained with some over-segmentation methods, and consists of two stages. In the first stage, the SAR image is classified with amplitude and texture feature used separately. Specifically, we use statistical model based maximum-likelihood method for amplitude based classification. Meanwhile, we classify the SAR image with the support vector machine (SVM) method by taking histograms generated with sparse coded morphological profiles as feature. To combine classification results produced with amplitude and texture features, a second refine stage is proposed based on the conditional random field (CRF) method. We define the CRF based on region adjacent graph (RAG) of superpixels. The unary term of the CRF is based on fusing classification scores produced by two classifiers in the first stage. Therefore, both of amplitude and texture information are used for the final classification. The graph cut (GC) algorithm is used to optimize the CRF model. We show experimental results on real SAR data, which verify the effectiveness of the proposed approach.
    2014 IEEE Radar Conference (RadarCon); 05/2014
  • Zongyong Cui · Zongjie Cao · Jianyu Yang · Jilan Feng
    [Show abstract] [Hide abstract]
    ABSTRACT: Synthetic aperture radar automatic target recognition (SAR ATR) has been widely applied in both military and civil fields. Much work has been done to improve the performance of SAR ATR systems in which feature extraction is an important step. To obtain pattern feature in SAR ATR, Non-negative Matrix Factorization (NMF), which is a dimensionality reduction method, has been applied by some researchers, although without deeper investigation. Meanwhile, in the computer vision field, lots of researches have been done to improve NMF methods by enforcing sparse constraint with L1-norm, like Non-negative Sparse Coding (NNSC), Local NMF (LNMF), and Sparse NMF (SNMF). Compared to L1-norm, L1/2-norm has been shown to have a more natural sparseness, however, little work has been done by using L1/2-norm constraint to NMF. In this letter, we propose a novel variant of NMF with L1/2 constraint, called L1/2-NMF, and carry out a thorough study by applying it in SAR target recognition. After mathematical derivation and analysis, the update rules of proposed L1/2-NMF are given in details. Experimental results on MSTAR public database show that both the basis and coding matrices obtained by L1/2-NMF have higher sparseness than those obtained by NMF, NNSC and NMF with Sparseness Constraints (NMFsc). The recognition results demonstrate that the proposed L1/2-NMF outperforms the other variants of NMF, like NNSC, NMFsc, and Nonsmooth NMF (NsNMF).
    2014 IEEE Radar Conference (RadarCon); 05/2014
  • Rui Min · Qianqian Yang · Yiming Pi · Zongjie Cao
    12/2013; 26(12):1069-1072. DOI:10.3724/SP.J.1187.2012.01069
  • Jifang Pei · Yulin Huang · Xian Liu · Jianyu Yang · Zongjie Cao · Bing Wang
    [Show abstract] [Hide abstract]
    ABSTRACT: Feature extraction from high-dimensional synthetic aperture radar (SAR) images is one of the crucial steps for SAR automatic target recognition (ATR). In this paper, we propose a new approach to SAR images feature extraction named Two-dimensional Principal Component Analysis-based Two-dimensional Maximum Interclass Distance Embedding (2DPCA-based 2DMIDE) which is based on manifold learning theory. The SAR image is projected into the feature space by horizontal 2DPCA and vertical 2DMIDE sequentially through this method. 2DPCA is efficient for image representation and preserves the global spatial structure of the original image, while 2DMIDE seeks to preserve the local spatial structure and the intrinsic geometry of the original image. Therefore, this feature extraction algorithm which fuses 2DPCA and 2DMIDE techniques can not only represent the original image in lower dimensions, but also excavate more powerful recognition information effectively. The experiment based on MSTAR database shows that the proposed method has a better recognition performance.
    2013 International Conference on Communications, Circuits and Systems (ICCCAS); 11/2013
  • Source
    Zongjie Cao · Ying Tan · Jilan Feng
    [Show abstract] [Hide abstract]
    ABSTRACT: This paper presents a variational model based segmentation approach for polarimetric synthetic aperture radar (PolSAR) images. The formulation for PolSAR image segmentation is based on a scaled Wishart distribution based continuous Potts model, which can partition the image domain into distinct regions with respect to the statistical property of PolSAR data. To make the segmentation efficient, a duality based optimization approach is utilized to minimize the energy functional. Moreover, an automatic initialization approach which takes the unsupervised H–a classification result of the polarimetric data as input is used to initialize the segmentation process. This approach can estimate the appropriate number of clusters and the corresponding classification map for the PolSAR data, which are used as the input of the following variational segmentation approach. In such a way, the proposed approach is carried out in a fully unsupervised way. Both of the polarimetric decomposition features and the statistical characteristics are used to get the final segmentation result, which helps to increase the accuracy. Experimental results demonstrate the effectiveness of the proposed approach. Without any artificial supervision, the proposed approach can produce superior segmentation results than results obtained with random initialized variational approach and Wishart–H–a classification approach.
    EURASIP Journal on Wireless Communications and Networking 01/2013; 2013(1). DOI:10.1186/1687-1499-2013-34 · 0.81 Impact Factor
  • Source
    Zongyong Cui · Zongjie Cao · Jianyu Yang · Jilan Feng
    [Show abstract] [Hide abstract]
    ABSTRACT: Synthetic aperture radar (SAR) automatic target recognition (ATR) is playing a very important role in military and civil field. Much work has been done to improve the performance of SAR ATR systems. It is well-known that ensemble methods can be used for improving prediction performance. Thus recognition using multiple classifiers fusion (MCF) has become a research hotspot in SAR ATR. Most current researchers focus on the fusion methods by parallel structure. However, such parallel structure has some disadvantages, such as large time consumption, features attribution conflict and low capability on confuser recognition. A hierarchical propelled strategy for multi-classifier fusion (HPSMCF) is proposed in this paper. The proposed HPSMCF has the characters both of series and parallel structure. Features can be used more effective and the recognition efficiency can be improved by extracting features and fusing the probabilistic outputs in a hierarchical propelled way. Meanwhile, the confuser recognition can be achieved by setting thresholds for the confidence in each level. Experiments on MSTAR public data demonstrate that the proposed HPSMCF is robust for variant recognition conditions. Compared with the parallel structure, HPSMCF has better performance both on time consumption and recognition rate.
    EURASIP Journal on Wireless Communications and Networking 01/2013; 2013(1). DOI:10.1186/1687-1499-2013-39 · 0.81 Impact Factor
  • Zongjie Cao · Lijia Chen
    [Show abstract] [Hide abstract]
    ABSTRACT: Because accurate identification cannot be obtained when the Identification Friend or Foe (IFF) sensor is employed separately, a radar sensor network (RSN) is designed to improve the identification capability in this paper. The content of this paper is focused on the information fusion algorithm, which is one of the key technologies in the RSN. The fuzzy c-means and the Bayesian network are chosen as the fusion algorithm. This algorithm can implement the identification friend or foe automatically after being trained by the training samples and expert's experience, and reduce the effect of uncertainties in the process of identification. At the same time, the algorithm can update the identification result with the augmentation of observations. The RSN can be expanded, if more information can be obtained, to adapt to the complicated environment, on the basis of this algorithm. The simulation results prove the validity and efficiency of the algorithm. Copyright © 2012 John Wiley & Sons, Ltd.
    Security and Communication Networks 06/2012; DOI:10.1002/sec.572 · 0.72 Impact Factor
  • Zongjie Cao · Zongyong Cui · Yong Fan · Qi Zhang
    [Show abstract] [Hide abstract]
    ABSTRACT: A hierarchical feature fusion strategy based on Support Vector Machine (SVM) and Dempster-Shafer Evidence Theory is proposed for SAR image automatic target recognition in this paper. This strategy has three fusion hierarchies corresponding to three features. Principle Component Analysis (PCA), Local Discriminant Embedding (LDE) and Non-negative Matrix Factor (NMF) features are extracted from images without preprocessing, and are fed to SVM classifier. However, not all features are used in each fusion process. At each fusion process, an empirical threshold T is used to determine the used features and hierarchy depth. Experiments on MSTAR public data set demonstrate that the proposed strategy outperforms the system combining the outputs of three features directly.
    Globecom Workshops (GC Wkshps), 2012 IEEE; 01/2012
  • Zongjie Cao · Jilan Feng · Rui Min · Yiming Pi
    [Show abstract] [Hide abstract]
    ABSTRACT: Feature extraction is a very important step in Synthetic Aperture Radar automatic target recognition (SAR ATR). In this paper, a feature extraction procedure based on the nonnegative matrix factorization (NMF) and Fisher linear discriminant (FLD) analysis is proposed for target recognition in SAR images. Firstly, segmented SAR images are processed by the NMF algorithm, which can extract nonnegative features that contain the local spatial structure information of targets. Then the FLD method is applied to the extracted features, thus the discriminability of the features can be enhanced. Both the spatial locality and separability between classes are enforced by this two-phase feature extracting procedure. Finally, the obtained features are used for automatic target recognition. Compared to several other methods, experimental results show the effectiveness of the proposed method for target feature extraction and recognition in SAR images.
    Communications (ICC), 2012 IEEE International Conference on; 01/2012
  • Qianqian Yang · Rui Min · Zongjie Cao · Yiming Pi
    [Show abstract] [Hide abstract]
    ABSTRACT: The Compressive Sensing (CS) theory, which was first promulgated by Candes and Donoho in 2004, can break through the Nyquist Sampling Theorem to solve the bottleneck problems on practice such as the high-speed sampling and the mass-data storing. In this paper, the CS idea is applied to multi-baseline SAR tomography 3-D imaging. Aiming at solving the bad imaging quality problem caused by sparse baselines in practical applications, the sparse model in the height direction of multi-baseline SAR is built according to the limited scattering centers hypothesis. Thus the problem of imaging in the height direction is transformed to the problem of reconstructing a sparse signal. And the lp-norm minimization methods are utilized to solve this problem. Simulation results show that the proposed method not only can overcome the limitation of Rayleigh Criterion, but also can solve the problem brought about by the non-uniform distribution of baselines in noisy environment.
    Globecom Workshops (GC Wkshps), 2012 IEEE; 01/2012
  • Xin Gao · Zongjie Cao · Yingxi Zheng · Yong Fan · Qi Zhang
    [Show abstract] [Hide abstract]
    ABSTRACT: Feature extraction is the key technology and the core task of Synthetic Aperture Radar (SAR) target recognition. In this paper, a new target feature extracting method based on Sparse Non-negative Matrix Factorization (SNMF) is presented, which mainly use SNMF as the method to decompose the SAR target image and to construct the sparse feature vector. By this means, the similarity inside each cluster of the feature vectors is improved and the difference between the clusters is also raised. An identification test using the classification method of Support Vector Machine (SVM) demonstrates that the proposed method, compared to PCA, ICA and the general NMF feature extraction methods, can improve the stability and the accuracy of the target recognition significantly.
    Globecom Workshops (GC Wkshps), 2012 IEEE; 01/2012
  • Rui Min · Yating Hu · Yiming Pi · Zongjie Cao
    [Show abstract] [Hide abstract]
    ABSTRACT: SAR Tomography Imaging Using Sparse Bayesian Learning
    IEICE Transactions on Communications 01/2012; 95-B(1):354-357. DOI:10.1587/transcom.E95.B.354 · 0.33 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: This paper focus on the problems coming from uncertainty and target's time variation in the identification friend or foe, and solves them by using an algorithm combining fuzzy c-means(FCM) and dynamic Bayesian network(DBN). The simulations prove the algorithm's validity.
  • Jianping Xu · Yiming Pi · Zongjie Cao
    [Show abstract] [Hide abstract]
    ABSTRACT: UWB linear frequency modulated (LFM) signals are widely used in radar, sonar and communication systems. In some applications, the detection of LFM signals and estimation of time-delay are very important. It needs very high sampling rate to address the problems for UWB LFM signal under Nyquist sampling theory which exceeds the current ADC capacity. In this paper, we propose a Compressive Sensing (CS) based method to solve the problem with ultra low sampling rate. We adopt an FrFt based sparse dictionary for CS because of the energy concentration property of LFM signal in the fractional Fourier domain. The performance is much better than the already existed method which used signal-matched sparse dictionary in noise condition. Experiments based on simulated data are carried out to testify the results.
    Signal Processing (ICSP), 2010 IEEE 10th International Conference on; 11/2010
  • Jilan Feng · Zongjie Cao · Yiming Pi
    [Show abstract] [Hide abstract]
    ABSTRACT: A variational level set approach based on the G0 distribution is proposed for SAR image segmentation problem. By introducing G0 distribution, the proposed method is more applicable for SAR image segmentation. To implement the segmentation, the parameters of the G0 distribution are estimated by the method based on Mellin transform, and level set function is applied for the numerical solution of PDEs which are derived from the minimization of the energy function. The experiments based on both synthetic and real SAR images prove that the proposed algorithm is applicable for a series of SAR images with different scenes.
    Synthetic Aperture Radar (EUSAR), 2010 8th European Conference on; 07/2010
  • Source
    Jianping Xu · Yiming Pi · Zongjie Cao
    [Show abstract] [Hide abstract]
    ABSTRACT: Compressive sensing (CS) is mainly concerned with low-coherence pairs, since the number of samples needed to recover the signal is proportional to the mutual coherence between projection matrix and sparsifying matrix. Until now, papers on CS always assume the projection matrix to be a random matrix. In this paper, aiming at minimizing the mutual coherence, a method is proposed to optimize the projection matrix. This method is based on equiangular tight frame (ETF) design because an ETF has minimum coherence. It is impossible to solve the problem exactly because of the complexity. Therefore, an alternating minimization type method is used to find a feasible solution. The optimally designed projection matrix can further reduce the necessary number of samples for recovery or improve the recovery accuracy. The proposed method demonstrates better performance than conventional optimization methods, which brings benefits to both basis pursuit and orthogonal matching pursuit.
    Journal on Advances in Signal Processing 02/2010; 2010. DOI:10.1155/2010/560349 · 0.81 Impact Factor
  • HongXia Zhang · ZongJie Cao · YiMing Pi
    [Show abstract] [Hide abstract]
    ABSTRACT: Most of the existing algorithms for multi-sensor tracks fusion are based on sensors with the same sampling rate, ignoring communication delay and different initial time. In this paper, a method to fuse information of asynchronous sensor tracks is proposed. At the track fusion center, position and velocity information of two sensors with different sampling rates and communication delays are fused. The fused track is an asynchronous track, which has minimum error covariance. This proposed fusion method not only avoids the time correction of asynchronous data, but also improves the tracking performance of multiple sensors. The performance of the proposed algorithm is also studied.