The paper proposes a new multiple-factor clustering method(NMFCM)with consideration of both box fractal dimension(BFD)and orientation of joints.This method assumes that the BFDs of different clusters were uneven,and c...The paper proposes a new multiple-factor clustering method(NMFCM)with consideration of both box fractal dimension(BFD)and orientation of joints.This method assumes that the BFDs of different clusters were uneven,and clustering was performed by redistributing the joints near the boundaries of clusters on a polar map to maximize an index for estimating the difference of the BFD(DBFD).Three main aspects were studied to develop the NMFCM.First,procedures of the NMFCM and reasonableness of assumptions were illustrated.Second,main factors affecting the NMFCM were investigated by numerical simulations with disk joint models.Finally,two different sections of a rock slope were studied to verify the practicability of the NMFCM.The results demonstrated that:(1)The NMFCM was practical and could effectively alleviate the problem of hard boundary during clustering;(2)The DBFD tended to increase after the improvement of clustering accuracy;(3)The improvement degree of the NMFCM clustering accuracy was mainly influenced by three parameters,namely,the number of clusters,number of redistributed joints,and total number of joints;and(4)The accuracy rate of clustering could be effectively improved by the NMFCM.展开更多
高光谱图像可以获取波段连续的图谱合一的立体数据,其具有丰富的图谱信息,能区分不同物质的类别,被广泛应用于各种遥感勘测领域。但在实际中高光谱图像的标注需要耗费大量的人力、财力和时间,可用的标注样本数量较少,难以通过训练来获...高光谱图像可以获取波段连续的图谱合一的立体数据,其具有丰富的图谱信息,能区分不同物质的类别,被广泛应用于各种遥感勘测领域。但在实际中高光谱图像的标注需要耗费大量的人力、财力和时间,可用的标注样本数量较少,难以通过训练来获得准确的分类结果,所以针对于只有少量标记样本的高光谱图像分类是一个挑战。近年来,自监督学习(Self-supervised Learning,SSL)已成为一种有效的方法,可以减少高光谱图像分类对昂贵的数据标注的依赖。SSL方法通过学习在同一图像的不同视图之间产生的潜在特征,在自然图像分类中取得了较高的分类精度。为了探索SSL方法在高光谱图像分类中的潜力,一种Bootstrap Your Own Latent(BYOL)框架下的自监督高光谱图像分类方法(BSSL)被提出。该方法通过引用自监督的图像特征学习框架BYOL,可以不需要负样本对,利用空间光谱相似的同类样本对进行网络训练及参数微调,提取到更具判别性特征。具体来说,该方法主要包括四个部分:BYOL的预训练、超像素聚类、基于“相似对”的BYOL的再训练和最终分类。为了验证该方法的有效性,在三个公开数据集上进行测试,并与五种先进的无监督、自监督分类方法SuperPCA、S3PCA、ContrastNet、SSCL和N2SSL进行对比,在Indian Pines和Salinas数据集上,BSSL方法的总体分类精度(OA)、平均分类精度(AA)、Kappa系数、召回率(recall)和f1分数(f1-score)都取得了更优值。其中在Indian Pines数据集上,OA分别比SuperPCA,S3PCA,ContrastNet,SSCL和N2SSL提高了1.32%,1.05%,5.68%,3.12%和1.27%。而在University of Pavia数据集上,BSSL方法表现没有那么出色,但在综合分类性能上也表现最优。这表明BSSL方法更适用于地物区域面积较大且分布较集中的场景,因为这对于超像素聚类来说更友好。展开更多
基金funded by the National Natural Science Foundation of China(Grant Nos.41972264 and 52078093)Liaoning Revitalization Talents Program,China(Grant No.XLYC1905015)。
文摘The paper proposes a new multiple-factor clustering method(NMFCM)with consideration of both box fractal dimension(BFD)and orientation of joints.This method assumes that the BFDs of different clusters were uneven,and clustering was performed by redistributing the joints near the boundaries of clusters on a polar map to maximize an index for estimating the difference of the BFD(DBFD).Three main aspects were studied to develop the NMFCM.First,procedures of the NMFCM and reasonableness of assumptions were illustrated.Second,main factors affecting the NMFCM were investigated by numerical simulations with disk joint models.Finally,two different sections of a rock slope were studied to verify the practicability of the NMFCM.The results demonstrated that:(1)The NMFCM was practical and could effectively alleviate the problem of hard boundary during clustering;(2)The DBFD tended to increase after the improvement of clustering accuracy;(3)The improvement degree of the NMFCM clustering accuracy was mainly influenced by three parameters,namely,the number of clusters,number of redistributed joints,and total number of joints;and(4)The accuracy rate of clustering could be effectively improved by the NMFCM.
文摘高光谱图像可以获取波段连续的图谱合一的立体数据,其具有丰富的图谱信息,能区分不同物质的类别,被广泛应用于各种遥感勘测领域。但在实际中高光谱图像的标注需要耗费大量的人力、财力和时间,可用的标注样本数量较少,难以通过训练来获得准确的分类结果,所以针对于只有少量标记样本的高光谱图像分类是一个挑战。近年来,自监督学习(Self-supervised Learning,SSL)已成为一种有效的方法,可以减少高光谱图像分类对昂贵的数据标注的依赖。SSL方法通过学习在同一图像的不同视图之间产生的潜在特征,在自然图像分类中取得了较高的分类精度。为了探索SSL方法在高光谱图像分类中的潜力,一种Bootstrap Your Own Latent(BYOL)框架下的自监督高光谱图像分类方法(BSSL)被提出。该方法通过引用自监督的图像特征学习框架BYOL,可以不需要负样本对,利用空间光谱相似的同类样本对进行网络训练及参数微调,提取到更具判别性特征。具体来说,该方法主要包括四个部分:BYOL的预训练、超像素聚类、基于“相似对”的BYOL的再训练和最终分类。为了验证该方法的有效性,在三个公开数据集上进行测试,并与五种先进的无监督、自监督分类方法SuperPCA、S3PCA、ContrastNet、SSCL和N2SSL进行对比,在Indian Pines和Salinas数据集上,BSSL方法的总体分类精度(OA)、平均分类精度(AA)、Kappa系数、召回率(recall)和f1分数(f1-score)都取得了更优值。其中在Indian Pines数据集上,OA分别比SuperPCA,S3PCA,ContrastNet,SSCL和N2SSL提高了1.32%,1.05%,5.68%,3.12%和1.27%。而在University of Pavia数据集上,BSSL方法表现没有那么出色,但在综合分类性能上也表现最优。这表明BSSL方法更适用于地物区域面积较大且分布较集中的场景,因为这对于超像素聚类来说更友好。
文摘针对超密集网络(ultra dense network,UDN)中基站密集部署导致的严重层间干扰问题,构建了考虑频谱复用和共信道干扰条件下最大化系统总吞吐量问题模型,提出了一种基于块坐标下降(block coordinate descent,BCD)法的联合频谱资源优化(joint resource optimization based on BCD,JROBB)方法。该方法将原问题分解为分簇、子信道分配和功率分配三个子问题,通过BCD法迭代优化子信道分配和功率分配,逼近原问题的最优解。仿真分析表明,在复杂度提升有限的情况下,系统总吞吐量比现有典型算法平均至少提升22%,可以有效提升频谱利用率。
