基于改进核聚类算法的空间目标识别方法

Space Target Recognition Based on Improved Kernel FCM

根据目标区域的矩特征,几何特征以及灰度特征,提取出目标的特征向量,并通过聚类算法对空间目标进行识别,提出了一种基于Voronoi距离的核聚类算法(KFCM)。该算法通过引入一种新的距离度量,使得隶属度函数更加的明晰,改善了核聚类算法极易陷入最小值的问题。运用改进的核聚类算法对3类空间目标进行识别,试验结果验证了算法的正确性和有效性。

A recognition system was presented to accomplish space target classification. Firstly, Image feature vectors were extracted according to invariant region moments, shape and image descriptors of space objects. And then, clustering algorithm is applied to the classify space target. An improved kernel clustering algorithm based on Voronoi distance was proposed which had a crisper membership function and was robust for noise and outliers. Experiments show that the improved kernel fuzzy clustering algorithm is more accurate and valid than that of the conventional methods.