一种新型月球地形自动识别迭代算法

A New Iterative Auto-Recognition Algorithm for Lunar Terrain

针对现有月球地形自动识别算法的识别率和精确度较低的问题,提出一种结合CCD图像和DEM数据信息,自动识别月球地形的动态分块迭代算法,实现了识别率和识别精度的双重提高。新算法提取CCD和DEM数据中月表地形的不同特征来构建图像子块的特征向量,再对特征向量聚类区分月球地形。算法根据输入图像的精度决定初始子块尺寸,提取子块的特征向量后聚类区分月海、月陆。每轮输出分类可信度高的子块结果后,会对分类结果可信度较低的子块进行细分,对细分后的子块重新提取特征向量并再次聚类分类,直到迭代算法终止。新算法已在三个典型的月面区域:虹湾(SI)、H010和危海(Crisium)区域进行了测试,试验结果与现有的地形分块识别算法相比,新算法的识别率和相关kappa系数均优于已知的自动识别算法结果。

In order to improve the recognition rate and accuracy of existing lunar terrain automatic recognition algorithm, we propose an iterative auto-recognition algorithm for lunar terrain which is based on CCD and DEM data. The extraction and analysis of lunar terrain features in this algorithm are based on both of CCD and DEM data, and each sub- block of image is clustered by the features of itself. Firstly, the algorithm determines the size of the initial block according to the accuracy of the input image. Then on the basis of the clustering result, directly output the classification results of block which has the high reliability classification, and continue to reduce the size of block which has low reliability classification results. And iteratively repeat extracting features and the result of clustering to distinguish the lunar mare and highland. Each round of iteration will output a number of classification results, and narrow the fuzzy area, until the algorithm terminates. This algorithm has been tested on three typical lunar surface areas： Sinus Iridum （SI）, HOIO and Crisium. Also the recognition rate and the Cohen＇s kappa coefficient of these areas have been calculated, and the results are better than the existing algorithms.