Dongliang Peng

A recursive algorithm for bearings-only tracking with signal time delay

For the bearings-only tracking (BOT) problem with signal propagation time delay, a recursive algorithm is proposed. Based on the analysis of signal time delay effect, an online parameter estimation (OPE) method is presented, which can be embedded in nonlinear filters to recursively estimate the target state with delayed measurements. The nonlinear filter used in this paper is a combination of the kernel-based filter (KBF) with an improved range parameterized (IRP) method, which manages subintervals according to their corresponding weights. The effectiveness and advantages of our proposed method compared with two existing methods are demonstrated by simulation examples.

Modeling of pH neutralization process using fuzzy recurrent neural network and DNA based NSGA-II

Abstract In this paper, the Takagi–Sugeno fuzzy recurrent neural network (T–S FRNN) is applied to model a pH neutralization process. Since the accuracy and complexity of the network are two contradictory criteria for the T–S FRNN model, a DNA based NSGA-II is proposed to optimize the parameters of the model. In the DNA based NSGA-II, each individual is encoded with one nucleotide base sequence, modified DNA based crossover and mutation operators are designed to improve the searching ability of the algorithm, and crowding tournament selection is applied based on the Pareto-optimal fitness and the crowding distance. The study on the performance of test functions shows that the DNA based NSGA-II outperforms NSGA-II in the quality of the obtained Pareto-optimal solution. To verify the effectiveness of the established T–S FRNN model for the pH neutralization process, it is compared with two T–S FRNN models optimized with other methods. Comparison results show that the model optimized by DNA based NSGA-II is more accurate and the complexity of the network is acceptable.

Augmented dimension algorithm based on sequential detection for maneuvering target tracking

In order to solve the problem that target tracking algorithm based on single model has poor tracking performance when the target occurs high maneuver and that IMM algorithm has low accuracy in tracking a constant velocity target, an augmented dimension algorithm based on sequential detection for maneuvering target tracking is proposed. First, the KF-UKF joint filtering is proposed. The Kalman filter based on the CV model is used to estimate the state of a constant velocity target. When the target maneuver is detected, the dimension of the CV model is augmented, and the unscented Kalman filter is used to estimate the state. Second, a fading memory sequential detection algorithm is proposed to detect the maneuver. Once the maneuver is detected, the augmented state vector and covariance matrix is compensated so that the modified model can match the actual motion mode. Simulation results show that this algorithm improves the accuracy of tracking by selecting the matching filter depending on the different mode of the target as well as modify the tracking state in real time.

A joint tracking and classification algorithm with improved mutual feedback

For the joint tracking and classification (JTC) problem in FM-band passive air surveillance radar system, a particle filter approach with improved mutual feedback is presented. Delay and Doppler measurements are used to estimate dynamic state and recognize target class. The improved mutual feedback between tracker and classifier is realized by a classification probability dependent particle assignment technique, which utilizes feedback information completely and increases tracking performance of the higher probability target class. Simulation results show the efficiency of the proposed method.

An Improved IMMJPDA Algorithm for Tracking Multiple Maneuvering Targets in Clutter

In this paper, the problem of tracking multiple maneuvering targets in clutter is investigated. An improved interacting multiple model joint probabilistic data association (IMMJPDA) algorithm is proposed. When the targets are described by different models, different association matrices are formulated. However, the traditional IMMJPDA algorithm only generates one association matrix. This new algorithm can reduce the overshoot of RMSE in position. The validity of this algorithm is illustrated through Monte Carlo simulations

SAR Imagery Scattering Center Extraction and Target Recognition Based on Scattering Center Model

Based on a scattering center parametric model derived from the geometric theory of diffraction, main characteristic scattering center Fisher optimal discriminator is presented in this paper for the problem of multi-class ground targets. All aspect main characteristic scattering center recognition model and target outline characteristic curve recognition model are presented in this paper. Four-class targets (Sandia Laboratories Implementation of Cylinders II, BTR70 armored transport, T72 main battle tank and BMP2 infantry tank) in MSTAR public database are adopted in our experiments and the results are presented

Applying differentiable mutual information to hyperspectral band selection

In this paper, we extend our earlier work by improving a mutual information (MI) based hyperspectral band selection method. Mutual information effectively measures the statistical dependence between two random variables. By modeling ground truth (e.g., a reference map) as one of the two random variables, MI can be used to find the spectral bands that contribute most to image classification. We apply a differentiable rather a histogram-based representation of mutual information to construct the estimated reference map, which results in an automatic solution by gradient searching. Experiments on the AVIRIS 92AV3C data set show that the proposed approach can find the best spectral window, and the bands in this window can be used to construct the reference map satisfactorily.

Band selection for hyperspectral image classification by a sliding window model

We investigate how to better use mutual information (MI) to select bands for hyperspectral image classification with less human intervention. Mutual information effectively measures the statistical dependence between two random variables. By modeling ground truth (e.g., a reference map) as one of the two random variables, MI can be used to find the spectral bands that contribute most to image classification. Extending our earlier work, we propose a sliding window model and apply mutual information to construct the estimated reference map, which need less human intervention. Experiments on the AVIRIS 92AV3C data set show that the proposed approach outperformed the benchmark methods, removing up to 55% of bands without significant loss of classification accuracy, compared to the 40% from that using the reference map accompanied with the data set. Meanwhile, its performance is found to be much robust to accuracy degradation when bands are cut off beyond 60%, revealing a better agreement in the mutual information estimation.

The sequential likelihood ratio test VSMM algorithm for maneuvering target tracking

The interacting multiple model state estimation approach is widely utilized for maneuvering target tracking. The IMM algorithm has a fixed model set structure which leads a dilemma that more models improve the accuracy but the use of too many models is as bad as that of too few models, and increases the computing burden. This paper presents a variable structure multiple model algorithm based on the sequential likelihood ratio test (SLRT-VSMM) that leads to a systematic treatment of model-set adaption. The new approach can increase the accuracy as well as decrease the computing burden by using the most likely model set at each time. The simulation of tracking an anti-ship missile shows the improvement of the system performance when we use the SLRT-VSMM approach.

An improved robust fusion method based on density estimation

To solve the uncertain information fusion problem, we define the robust performance of the fusion algorithm and propose an improved density estimation based robust fusion algorithm. First, the mean-shift procedure is employed in detecting the dominant mode of the information density function; second, the max iterative time is calculated according to the real time request; last, all the valid data in the dominant field after the max iterative time are fused. The presented algorithm is compared with the weighted fusion method and the density estimation fusion technique in two simulation cases. The results show that it is more accurate and robust and satisfies the system real time request.