Guo Guirong

An intelligence recognition method of ship targets

The problem of ship target recognition is discussed, and an intelligent recognition method is proposed. An experimental system constructed on the basis of the method consists of a high-rate waveform recorder, a high-speed signal-processing system, a data processor, and a special interface. An experimental study of ship target recognition has been carried out on an actual low-resolution radar. The experimental results show an overall success rate of more than 90% when there are six classes of ships.<<ETX>>

Automatic HRR target recognition based on Prony model wavelet and probability neural network

An automatic high range resolution (HRR) target recognition algorithm is detailed and tested on a data set of five different aircraft. A super-resolution downrange profile of radar returns of HRR is obtained using the Prony model. Target features are extracted by the wavelet transform. The features consist of two parts: one reflects the detailed structure of the targets, the other shows the outline of the targets. A probabilistic neural network (PNN) with a simple data fusion technique is applied for target classification.

A new data extrapolation algorithm with application in guidance and target recognition

A new data extrapolation algorithm is presented for guidance and target recognition application. The algorithm firstly uses DFT or FFT computation of L points data of response over band (f/sub o/,f/sub o/+/spl Delta/F) to obtain the rough outline of target range profile. Then a CFAR detector is used to obtain the structure window function X/sub w/(e/sup j/spl omega) and its IFT series X/sub w/(l) and several filtering functions {H/sub j/(z)} are constructed. Finally, the extrapolation process by using the filtered series {y/sub j/(l)} based on X/sub w/(e/sup j/spl omega) restriction is established. The algorithm has the advantage of robustness under clutter background.<<ETX>>

Target discrimination and recognition using high resolution range features

A method for target discrimination and recognition based on target high-resolution range profiles (HRRPs) is summarized. This method uses heuristic feature extractions and the Mellin transformation of HRRPs. The use of wideband millimeter waves leads to better discrimination and recognition performance. The key techniques involved are: using wideband waveform to stimulate target scattering centers information; using a very-high-speed one- or two-dimensional FFTR processor to obtain the range profile or velocity of the target; and synthesis, fusion, and transformation of range profile features.<<ETX>>

Relationship between radar target signatures and motion modes

Aerial and ground targets motion modes have a close relationship with their pose angles, hence with radar signatures. An analysis on target pose angular rates under different motion modes is performed firstly. Expressions of radar signatures are then derived, including radar cross section (RCS), angular glint errors, and high resolution Doppler profile. By taking a two-scatterer simple target as an example, we explore the relationship between the above signatures and target pose angular rates. The relationship between radar signatures and motion modes are then developed. Finally, the simulation results of radar signatures for multiple-scatterer complex target validate the theoretical analysis, and show that it is feasible to identify different motion modes based on radar signatures. The conclusions are significant for the research of motion mode identification, maneuver detection, and maneuvering target tracking and interception using radar signatures.

Approach to the fractal features of high-resolution polarimetric radar targets

For high-resolution radar, a novel idea is to use the polarimetric range profiles understanding technique to improve recognition performance. In this paper, a study is performed based on the fractal characterisation of one-dimensional range profiles of high-resolution polarimetric radar targets at four kinds of polarimetric states (HH, HV, VH, VV), and a new concept of the fractal matrix is proposed.

Measuring evidential consistence by a generalized relative entropy

The overall objective of this paper is to derive an algorithmic framework for measuring the evidential consensus of Dempster-Shafer theory. More specifically, our goal is to develop information theoretic criteria to quantitatively measure the quality of consensus generated by pooling evidences from multi-knowledge sources.

Correlation methods for aircraft identification from fully polarized radar range profiles

There is a variety of ways to partially characterize an aircraft using radar. A very simple and rapid way to characterize an aircraft is through the use of high resolution range profiles. Hudson and Psaltis (1993) investigate the use of real radar range profiles for identification purposes, they concentrate on a correlator-based system, and achieve good results. In this paper, we extend their work to fully polarized situation. The results of our experiments show that correct identification rates of five scale-models aircraft are 100%, 95%, 92%, 100%, 100% and suggest that reliable aircraft identification is possible provided estimated aspect information and fully polarized profiles are used.<<ETX>>

The neural network method for radar weak target detection

Because of the statistical nature nature of many types of clutter, a radar target detector must set a fairly high threshold in order to order to maintain a reasonable false-alarm rate. However, weak targets are usually missed for the above threshold detector. This paper presents an effective detector, which can be considered as a two-dimensional feature matching filter for radar signals. The feature extraction is performed by Hopfield neural networks and the feature integration is finished by a multilayer perceptron. In order to overcome the local optimum problem, a novel modification which is called energy comparing method is introduced into the Hopfield model dynamic equation to find the global optimum. By testing with the real radar return data in a low signal-to-clutter ratio, the detector presented in this paper has more advantages than the conventional threshold detector.<<ETX>>

Fuzzy sets-based neural network for pattern understanding

A fuzzy classification process model and a rational neural network topology are suggested and studied. A new method of constructing membership functions is proposed by using a self-organizing feature map network, kernel estimation of the probability distribution, and a consistent transformation between probability and possibility. Sugenos (1974) fuzzy integral is briefly reviewed. Then, an improved fuzzy integral, which is based on double set measures, is proposed. The corresponding classification neural network is underlined and analyzed. This fuzzy set-based neural network can combine fact-level information with knowledge-level information consistently, and its classification process is almost identical to the human cognitive process. The given test results show that simultaneously high levels of robustness and accuracy for radar ship classification have been reached by using the proposed fuzzy set-based neural network.<<ETX>>