基于Fukunaga-Koontz变换的高光谱图像异常检测

Fukunaga-Koontz Transform Based Anomaly Detection Algorithm for Hyperspectral Imagery

高光谱图像异常检测算法通常是基于数据变换,在新的特征空间中进行的。针对WDEST算法没有使得异常目标和背景在新的特征空间中有较好的分离,提出了一种基于Fukunaga-Koontz(FKT)变换的高光谱图像异常检测算法。该算法利用FKT对高光谱图像局部窗口中数据进行变换,使得在新的特征空间中异常目标和背景有相同的特征向量和互补的特征值,较之WDEST算法得到了更好的分离,在有效提高检测概率的同时降低了虚警概率;经与RX算法比较表明,该算法对于较大异常有更好的适应性,并用真实数据进行实验证明了算法的有效性。

Anomaly detection algorithms are commonly based on transforming data into a new space to implement for hyperspectral imagery. Aiming at WDEST algorithm failed to discriminate anomaly and background in a new space, a new method based Fukunaga-Koontz transform （FKT） is proposed. Utilizing FKT to transform data in local windows into a new space, anomaly and background have same eigenvectors and complimentary eigenvalues. Compared to WDEST, a higher degree of discrimination of anomaly and background is generated and results in a reduction in both the miss rate and the obvious false alarm rate. In addition, the proposed approach has better adaptive to larger anomaly than RX algorithm. The effectiveness of the proposed method is validated by experimental results obtained from real data.