Wang Wenguang

Oil spill detection from polarimetric SAR image

A new combined feature is proposed based on polarimetric features extraction. Moreover, a new oil spills detection method is developed based on the combined feature and max entropy segmentation. The application to NASA/JPL SIR-C data shows that the new combined feature and detection method are effective to oil spills detection.

River detection from SAR images

River detection from SAR images plays an important role in civilian applications. A new method for river detection is proposed in this paper, which includes fuzzy clustering, wavelet transform and using snake model. River area can be extracted by clustering and morphological processing. Then the edge of river is extracted with the wavelet modulus maximum method (WTMM), and is smoothed by the snake model. A Radarsat-1 image is used for the experiment. The experimental result shows that the method proposed in this paper is efficient for river detection and edge location.

Research of Small Target Detection within Sea Clutter Based on Chaos

In this paper, based on the prior knowledge of chaotic character of sea clutter, we discuss a method for detection of small target in sea clutter using neural network as a predictor. Neural network can capture the dynamics of strange attractor generating sea clutter. After the reconstruction of real-life Radar data, BP and RBF networks are regularized, and then a set of sample data from real-life data is inputted into the networks to train the neural networks respectively. As a sequence, according the nature of neural networks, these trained networks could approximate the model of dynamical system responsible for sea clutter. These networks can be used to detect small target within sea clutter as a predictor.

Recognition of SAR image based on combined templates

In this paper, a new recognition algorithm of SAR image which is based on combined templates has been proposed. The new algorithm is based on the traditional mean templates recognition. We use some statistical information of the training samples to make a refusing threshold, which is expected to have the ability that can refuse the non-template-class targets effectively. Meanwhile, the proposed combined templates algorithm can re-recognize those refused targets by another target recognition method called recognition based on features of graphic image knowledge so that we can improve the final recognizing efficiency. Finally, we use MSTAR database to detect the target recognition ability, which can indicate that the algorithm we proposed is effective.

A novel CFAR detector in heterogeneous environment

In nonuniform situations, training data may not be representative of the disturbance in the Cell under Test (CUT) and the CA-CFAR exhibits strong degradations both in the detection performance as well as in the CFAR behaviour. Therefore a GIS-based training data selector and a statistical clutter edge selector are used to obtain homogeneous clutter field before the CFAR process. The simulation result in the presence of real data shows that the proposed CFAR detector not only keeps lower number of false alarms, but also improves the detection performance of classical CFAR detector by taking use of the environmental context and the clutter field parameters.

Urban construction area extraction using circular polarimetric correlation coefficient

Urban construction area detection is of great significance for tracking, mission planning, training, loss estimation and urban planning. In this paper, we make full use of the polarization characteristics of SAR (synthetic aperture radar) data to detect urban construction area. First, circular polarization correlation coefficient characteristics, entropy characteristics based on the gray level co-occurrence matrix (GLCM), and the dihedral angle scattering characteristics using the Pauli decomposition are extracted to distinguish among urban area, forest area and other manmade targets. And then we adopt the three kinds of characteristic to form feature vector and complete urban area detection based on k-means clustering analysis. The experimental result has proved the efficiency of this method.