改进模糊聚类语义分割声环境功能区划图  

Semantic Segment of Acoustic Environment Functional Zoning MapBased on Improved Fuzzy Clustering

作　　者：曾宇[1] 姚琨 秦勤 ZENG Yu;YAO Kun;QIN Qin(Institute of Urban Safety and Environmental Science,Beijing Academy of Science and Technology,Beijing 100054,China)

机构地区：[1]北京市科学技术研究院城市安全与环境科学研究所,北京100054

出　　处：《噪声与振动控制》2025年第2期210-215,共6页Noise and Vibration Control

基　　金：北京市科学技术研究院城市安全与环境科学研究所任务型科研课题(SZ-SKT-220346)。

摘　　要：声环境功能区划多采用地理信息系统进行研究,但公开发布的声环境功能区划方案中的文字和图片无法直接用于地理信息系统分析。首先提出改进模糊C均值聚类超像素方法,对声环境功能区划图进行语义分割以获取声功能区信息。接着采用简单线性迭代聚类构建超像素,提取声环境功能区划图特征矩阵,基于K-means++改进模糊C均值聚类算法,语义分割超像素粒化的声环境功能区划图,并以声功能区面积占比计算结果偏差为评价指标,分析超像素尺度对分割结果的影响。然后基于不同图像特征矩阵构建方法和聚类中心初始化方法,使用模糊C均值聚类、高斯混合模型聚类、K-medoids聚类语义分割声环境功能区划图,最后比较不同组合方案的声功能区面积占比计算结果偏差,验证方法的有效性。Acoustic environment functional zoning is usually studied using Geographic Information Systems. However,the text and images in publicly released acoustic environment functional zoning schemes cannot be directly used for GeographicInformation System analysis. This paper proposed an improved Fuzzy C-Means clustering superpixel method. Firstly,the acoustic functional zone information was acquired by semantic segmentation of acoustic environment functional zoningmaps. Then, the map′ s superpixels were constructed by simple linear iterative clustering to extract the map′ s characteristicmatrix, and the superpixel-granulated map was segmented based on improved K-means++ Fuzzy C-Means Clustering.With the deviation between the calculated area proportion of acoustic functional zones as the evaluation index, the influenceof superpixel scale on the segmentation results was analyzed. Afterwards, based on different image characteristic matrix constructionmethods and cluster center initialization method, and using Fuzzy C-Means Clustering, Gauss hybrid model clusteringand K-modoids clustering, the semantic segmentation of the acoustic environment functional zoning map was realized.Finally, deviation between the calculated area proportion of acoustic functional zones of different semantic segmentationschemes based on different clustering methods combination were analyzed and compared , which verified the validity of theproposed method.

关 键 词：声学 声环境功能区划图 彩色图像分割 模糊C均值聚类 简单线性迭代聚类 K-means++算法 

分 类 号：X827[环境科学与工程—环境工程] TP391[自动化与计算机技术—计算机应用技术]