采用目标背景建模的毫米波弱小目标检测

基于被动毫米波成像特性,提出了改进的稀疏表示——圆周中心差(ISR-CSCD)算法来解决被动毫米波图像中弱小目标与背景区分度较弱,目标可提取特征较少的问题。该算法通过改进稀疏表示方法完成背景抑制与目标增强。依据目标与周围背景特征先验,提出了圆周中心差背景抑制算法对检测图像进行背景抑制。然后,融合改进稀疏表示方法和圆周中心差背景抑制算法的结果得到抑制了背景的目标增强图像。最后,基于恒虚警率的检测方法完成了弱小目标的检测。对不同场景下的毫米波图像进行了实验检测,结果表明,与主流算法图像稀疏表示(SR)法、鲁棒规则核回归牛顿算法(NRRKR),空时联合分类稀疏表示算法(STCSR)和累积中心与周边差异测量算法(ACSDM)相比,ISR-CSCD算法具有更低的虚警率、更高的检测精度、更强的鲁棒性。对各种虚警率、信噪比之下的毫米波弱小目标检测结果显示,ISR-CSCD检测率相对于其它算法平均提高了约15%。

Millimeter wave dim small target detection based on target and background modeling

On the basis of characteristics of Passive Millimeter Wave (PMMW) imaging, an Improved Sparse Representation-Circle-Surround Center Difference(ISR-CSCD) algorithm is proposed to improve the weaker distinction between dim small target and background and the smaller target features to be extracted. The algorithm firstly improves the sparse representation to complete the background suppression and target enhancement. Then, according to the features and prior knowledge of the target and the surrounding background, the background suppression algorithm of circle-surround center difference is used to suppress the background of the image. The results by two methods mentioned above are fused to get the final enhanced target image. Finally, the Constant False Alarm Rate (CFAR) is used to complete dim small target extraction. The millimeter wave images in different scenes are detected. The results show that as compared with the mainstream algorithms, Sparse representation (SR), Newton methods for Robust Regularized Kernel Regression(NRRKR), Spatio-temporal Classification Sparse Representation(STCSR) and Accumulated Center-surround Difference Measurement(ACSDM), the ISR-CSCD algorithm has a lower false alarm rate, higher detection accuracy and stronger robustness. For all kinds of false alarm rates and the signal to noise ratios of the millimeter wave small target detection results in statistics, the detection rate of ISR-CSCD is increased by about 15% as compared with other algorithms.