论文部分内容阅读
本文针对高分辨率遥感影像快速高效萃取有用信息这一遥感技术应用的热点问题,探讨了一种适合于组合特征识别的遥感图像最近邻模糊分类器。该分类器首先把待识别目标的组合特征与训练模板中的组合特征样本的平均值一一进行比较,从而得到了一个特征差矩阵。用模糊分布函数在同类特征差之间进行处理,生成一个隶属度矩阵,然后用算术平均法对隶属度矩阵进行处理,并用最大隶属度准则来进行分类判决。以新疆和静县的SPOT5图像为例,应用此方法对其进行分类试验。结果表明:利用此分类方法对SPOT5遥感图像进行分类,不仅使分类结果具有丰富的语义信息,而且克服了由于特征选择的不稳定性对分类结果的影响,分类精度也得到了显著的提高。
In this paper, aiming at the hot issues in remote sensing application of high-resolution remote sensing images, this paper discusses a nearest neighbor fuzzy classifier of remote sensing images which is suitable for combined feature recognition. The classifier first compares the combined features of the target to be recognized and the average of the combined feature samples in the training template one by one to obtain a feature difference matrix. The fuzzy distribution function is used to deal with the differences between similar features to generate a membership matrix. Then, the membership degree matrix is processed with the arithmetic average method, and the classification decision is made with the maximum membership degree criterion. Taking the SPOT5 image in Hejing county of Xinjiang as an example, this method was used to classify it. The results show that using this classification method to classify the SPOT5 remote sensing images not only makes the classification results rich in semantic information, but also overcomes the influence of the instability of feature selection on the classification results and improves the classification accuracy significantly.