论文部分内容阅读
针对ISODATA算法预设参数较多,其聚类中心与最优迭代数目很难预先准确设定,且在聚类时没有将影像自身特点充分考虑,对个体适应度函数重视不够的问题,本文提出一种融合增强型模糊聚类GA与ISODATA的聚类方法,对聚类原型矩阵进行编码,构造隶属度矩阵,解求个体适应度函数值,在影像特征空间中搜索得到样本全局收敛极值点。通过试验证明,该方法能避开随机初选值的敏感问题,避免聚类过程的随机性,使分类结果与实际情况更为接近,该算法精度优于传统的ISODATA算法与模糊聚类GA算法,提高了分类的精度,整体效果较好。
For the ISODATA algorithm, there are many preset parameters, the clustering centers and the optimal number of iterations are difficult to set in advance, and the clustering does not fully consider the features of the image and the problem of individual fitness function is not enough. A clustering method based on GA and ISODATA is proposed. The clustering prototype matrix is encoded and the membership degree matrix is constructed. The fitness value of individuals is solved and the global convergence of the sample is obtained by searching the image feature space . The experiment proves that this method can avoid the sensitive problem of random primaries and avoids the randomness of the clustering process so that the classification result is closer to the actual situation. The accuracy of this algorithm is better than the traditional ISODATA algorithm and fuzzy clustering GA algorithm , Improve the classification accuracy, the overall effect is better.