中文摘要：

在复杂背景干扰下,高光谱图像异常检测虚警率较高。针对这一问题,提出了结合非下采样contourlet变换（nonsubsampled contourlet transform,NSCT）和空间聚类的基于支持向量数据描述（support voctor data description,SVDD）的异常检测算法。首先通过对高光谱数据进行NSCT分解,得到含有绝大部分背景信息的低频图像,与原始图像进行差运算,获取背景残差图像,以此抑制背景信息的干扰; 然后采用空间聚类法对低频图像进行聚类分割,获得各子区域的特征光谱作为SVDD训练样本进行背景建模,克服异常像元与图像随机噪声对SVDD背景建模的影响,同时降低计算量; 最后利用得到的SVDD模型对背景残差图像进行异常检测。实验结果表明,算法抑制了复杂背景的干扰,降低了虚警率,更适用于高光谱图像全局异常检测。

英文摘要：

Due to the interference of complex background information, anomaly detection algorithm has incremental false alarm rate. In order to overcome this problem, this paper proposes an improved SVDD algorithm combining the nonsubsampled contourlet transform （NSCT） with spatial clustering. Hyperspectral imagery is transformed by NSCT, and the low frequency image containing most background information is obtained. The background residual error which is the minus of the hyperspectral imagery and low frequency image can be acquired, whereupon the background information is suppressed. Then, the low frequency image is clustered by spatial clustering method, thereupon the feature spectrum of each sub-region is computed and used as a training sample for SVDD. Hence it can eliminate the influence induced by the anomalous spectrum or random noise, and the calculated amount is also reduced at the same time. Finally, the SVDD model is used to detect background residual error data. The results show that the proposed method can inhibit the interference of complex background. It has lower false alarm rate, and hence it is more appropriate for global anomaly detection in hyperspectral imagery.