采用多分辨率分解的高光谱图像异常检测

Anomaly detection for hyperspectral image based on multiresolution decomposition

针对异常检测中背景杂波的干扰问题,提出了一种基于多分辨率分解的异常检测方法。首先对高光谱图像进行分解,将其划分为不同频率的子块;其次,构造核函数来抑制背景杂波;最后将检测算子应用于处理后的子块,重构图像,检测出异常目标。由于图像被分解到不同频带,子块中包含的背景信息大大减少,此时对子块进行检测,很好地削弱了背景杂波对搜索异常信号的干扰;此外,使用构造的核函数对子块中的背景进行抑制,抑制后的异常信号远离背景,更易检出异常目标。实验结果进一步验证了该方法具有很好的异常检测性能。

According to the interference of background clutter on the anomaly detection,a novel anomaly detection method for hyperspectral images based on multi-resolution decomposition was presented.Firstly,the hyperspectral image was decomposed into a series of different frequency sub-bands.Secondly,the background clutter of the sub-bands were suppressed by the structured kernel function.Finally,the detection operator was used in the sub-bands,and the detection result was obtained by reconstructing the sub-bands.The background information was greatly decreased due to the process of multi-resolution decomposition,following which the interference of the background clutter was well weakened for the anomaly signal detection by use of the detection operator.Moreover,after suppressing the background clutter using the structured kernel function,the anomaly signals get far away from the background and can be detected more easily.The experimental results prove the validity of this method.