基于二级聚类的遥感影像变化检测  

Remote sensing images change detection based on secondary clustering

作　　者：范婕 贾付文 武昌东 FAN Jie;JIA Fuwen;WU Changdong(College of Computer and Information Technology,Three Gorges University,Yichang Hubei 443002,China Hubei Engineering Technology Research Center for Farmland Environment Monitoring,Three Gorges University)

机构地区：[1]三峡大学计算机与信息学院三峡大学湖北省农田环境监测工程技术研究中心,湖北宜昌443002

出　　处：《激光杂志》2023年第8期49-53,共5页Laser Journal

基　　金：国家自然科学基金青年项目(No.41901341)。

摘　　要：为了提高现有基于二级聚类变化检测方法中第一级聚类结果的准确性以及解决第二级聚类时分类器计算复杂度高、耗时久的问题,提出一种基于引入信息熵的模糊C均值(Fuzzy C-means,FCM)算法和K近邻(K-Nearest Neighbor,KNN)算法级联的变化检测方法。首先通过FCM算法对差分影像(Differential imagery,DI)聚类并引入信息熵计算出DI中每个像元的不确定性,通过分析不确定性强弱得到更为可靠的训练样本,第二级聚类时使用计算复杂度更低的KNN算法代替深度学习方法进行分类得到最终变化检测结果。三组真实遥感影像数据集上的实验结果表明该方法能够有效在降低计算复杂度的同时提高变化检测性能:在Sulzberger数据集上的Kappa系数为96.83%,比其他方法提高2.69%~4.55%;在Madeirinha1数据集上的Kappa系数为85.69%,比其他方法提高0.18%~1.99%;在Madeirinha2数据集上的Kappa系数为87.47%,比其他方法提高0.20%~3.53%。In order to improve the accuracy of the first-level clustering results in the existing second-level cluste-ring-based change detection method and solve the problem of high complexity and time-consuming computation of the classifier during two-level clustering,a cascade change detection method based on the fuzzy C-means(FCM)algo-rithm and K-Nearest Neighbor(KNN)algorithm based on the introduction of information entropy is proposed.The Differential imagery(DI)is clustered by the FCM algorithm and the information entropy is introduced to calculate the uncertainty of each pixel in DI,and more reliable training samples are obtained according to the uncertainty strength,and the KNN algorithm with lower computational complexity is used instead of the deep learning method to classify the final change detection results in the second level of clustering.Experimental results on three sets of real remote sensing image datasets show that the proposed method can effectively improve the change detection performance while reducing the computational complexity:the Kappa coefficient on the Sulzberger dataset is 96.83%,which is 2.69%~4.55%higher than other methods.The Kappa coefficient on the Madeirinha1 dataset was 85.69%,which was 0.18%~1.99%higher than other methods.The Kappa coefficient on the Madeirinha2 dataset is 87.47%,which is 0.20%~3.53%higher than other methods.

关 键 词：二级聚类 信息熵 遥感影像 变化检测 

分 类 号：TN29[电子电信—物理电子学]