MSTAR Database Research Articles

Recognition method of multi-feature fusion based on D-S evidence theory in SAR image

In view of the low accuracy of the single feature-based method for target recognition in SAR image,a multi-feature decision-making level fusion method based on SVM and D-S evidence theory was proposed.After a series of image processing,the texture feature,Hu invariant moments feature and peek feature were extracted from the target image.Then the targets were classified according to each type of features utilizing SVM,and the results were used as evidence to construct the basic probability assignment.Conclusively,D-S combination rule of evidence was used to achieve fusion,and final recognition results were given by classification thresholds.The method is used for recognizing three-class targets in MSTAR database,and the recognition rate arrives at 95.5%.Experimental result shows that the method is effective for SAR images target recognition.

TARGET RECOGNITION FOR MULTI-ASPECT SAR IMAGES WITH FUSION STRATEGIES

Two fusion strategies for target recognition using multi- aspect synthetic aperture radar (SAR) images are presented for recognizing ground vehicles in MSTAR database. Due to radar cross- section variability, the ability to discriminate between targets varies greatly with target aspect. Multi-aspect images of a given target are used to support recognition. In this paper, two fusion strategies for target recognition using multi-aspect SAR images are proposed, which are data fusion strategy and decision fusion strategy. The recognition performance sensitivity to the number of images and the aspect separations is analyzed for those two target recognition strategies. The two strategies are also compared with each other in probability of correct classiflcation and operating e-ciency. The experimental results indicate that if we have a small number of multi-aspect images of a target and the aspect separations between those images are proper, the probability of correct classiflcation obtained by the two proposed strategies can be advanced signiflcantly compared with that obtained by the method using single image.

A MULTI-SCALE LOCAL PHASE QUANTIZATION PLUS BIOMIMETIC PATTERN RECOGNITION METHOD FOR SAR AUTOMATIC TARGET RECOGNITION

Synthetic aperture radar (SAR) automatic target recog- nition (ATR) has been receiving more and more attention in the past two decades. But the problem of how to overcome SAR target am- biguities and azimuth angle variations has still left unsolved. In this paper, a multi-scale local phase quantization plus biomimetic pattern recognition (BPR) method is presented to solve these two di-culties. By applying multiple scales local phase quantization (LPQ) on the ob- served SAR images, the blur and azimuth invariant features can be extracted, and these features are fusion at consecutive multiple scales to achieve better results. Then PCA method is applied to further re- duce the feature dimension and achieve its e-ciency. Finally, high dimensional space geometry covering method based on BPR theory is adopted to construct hyper sausage neuron links for target recognition. Experiments on the MSTAR database show that the proposed method can achieve satisfying recognition accuracy compared with other state- of-the-art methods.

Target Recognition of Synthetic Aperture Radar Images Using Level-Set Segmentation Method

Correct segmentation and effective feature extraction is a critical problem in target recognition of synthetic aperture radar images.In this paper,level-set method was applied in target recognition of synthetic aperture radar images.Target and shadow regions of each image were segmented from background clutter regions using level-set method.The shape of target region was used as significant feature in target recognition based on template matching.Image samples of three targets in MSTAR database are used to verify the method.The results show that using level-set segmentation and template matching can extract effective feature of target and achieve correct recognition of targets.

CFAR Detection of Extended Targets in SAR Images Based on Goodness-of-Fit Test

A new constant false alarm rate (CFAR) detection scheme based on the goodness-of-fit (GoF) test is proposed to deal with the problem of extended object detection in high resolution synthetic aperture radar (SAR) images. The performance of this detector is compared with that of the traditional detectors using the MSTAR database. Results show that the proposed detector is superior to the traditional detectors in controlling false alarms in nonhomogeneous environment where boundaries widely exist.

Persistence modeling of angularly dependent synthetic aperture radar imagery

An image persistence framework is developed to analyze azimuthally varying synthetic aperture radar (SAR) data. The model focuses on cases containing rich aspect data from a single depression angle. The goal is to replace the data's intrinsic viewing geometry dependencies with target-specific dependencies. Both direct mapping functions and cost functions are presented for data transformation. An intensity-only mapping function is realized to illustrate the persistence model in terms of a canonical example, visualization, and classification. The limitations of an intensity-only mapping function are also discussed. Because the new image-space output from the persistence model is closely tied to the radio frequency (RF) characteristics of two different targets and not to the collection geometry, it shows promise for integration into various automated target recognition (ATR) algorithms. Target classification potential was tested using the MSTAR database and simplified template matching schemes. The intensity-only transformation function permits a very low number of persistence images to represent the general RF target characteristics of an assortment of vehicles. The ability to represent target characteristics with relatively low resources could also be beneficial in ATR applications. Overall, the persistence framework shows strong potential as a new tool that can be used in the analysis of multiaspect SAR images.

Pose estimation of SAR imagery using the two dimensional continuous wavelet transform

We develop a novel pose estimation technique for synthetic aperture radar image chips based on the 2-D continuous wavelet transform (CWT). The pose estimator exploits a fast approximation of the 2-D CWT so that its computational complexity is comparable to a principle component analysis (PCA) based approach. By incorporating energy from various scales, the CWT approach is more robust than the PCA method. Experiments over the MSTAR database confirm the robustness of the CWT method.

Target Recognition Based on Directional Filter Banks and Higher-Order Neural Networks

Park, Sang-il, Smith, Mark J. T., and Mersereau, Russell M., Target Recognition Based on Directional Filter Banks and Higher-Order Neural Networks, Digital Signal Processing10 (2000), 297–308.A new approach for the classification of SAR targets is presented here, which combines maximally decimated directional filter banks with higher-order neural networks (HONNs). HONNs are neural networks that can achieve performance similar to that of standard multilayered neural networks, but without the hidden layer. Their performance can be made invariant to geometric transformations of the input imagery in the design process, while their computational complexity can be reduced by employing a preprocessor to reduce the dimensionality, such as coarse coding. Most past image classifiers using HONNs have been designed for carefully thresholded binary images. However, generating useful binary representations that can be used as inputs can be difficult for modalities such as SAR. As an alternative, we use a novel HONN implementation that accepts gray-level input pixels using directional filter banks. In order to do this, a new modified tree-structured directional filter bank structure in a very computationally efficient form is introduced. The performance of the proposed approach is demonstrated and compared in imagery taken from the public MSTAR database. The new approach is shown to be effective in enhancing the discrimination power of the HONN inputs, leading to significantly improved performance.

Synthetic aperture radar automatic target recognition with three strategies of learning and representation

We describe a new architecture for synthetic aperture radar (SAR) automatic target recognition (ATR) based on the premise that the pose of the target is estimated within a high degree of precision. The advantage of our classifier design is that the input space complexity is decreased with the pose information, which enables fewer features to classify targets with a higher degree of accuracy. Moreover, the training of the classifier can be done discriminantly, which also improves performance and decreases the complexity of the classifier. Three strategies of learning and representation to build the pattern space and discriminant functions are compared: Vapnik's support vector machine (SVM), a newly developed quadratic mutual information (QMI) cost function for neural networks, and a principal component analysis extended recently with multiresolution (PCA-M). Experimental results obtained in the MSTAR database show that the performance of our classifiers is better than that of standard template matching in the same dataset. We also rate the quality of the classifiers for detection using confusers, and show significant improvement in rejection.