建筑物多边形高精度识别的傅里叶形状描述子神经网络方法  

Fourier descriptor-based neural network method for high-precision shape recognition of building polygon

作　　者：刘鹏程[1,2] 黄欣[1,2] 马宏然 杨敏 LIU Pengcheng;HUANG Xin;MA Hongran;YANG Min(Key Laboratory for Geographical Process Analysis&Simulation of Hubei Province,Central China Normal University,Wuhan 430079,China;School of Urban and Environmental Sciences,Central China Normal University,Wuhan 430079,China;School of Resource and Environment Sciences,Wuhan University,Wuhan 430079,China)

机构地区：[1]华中师范大学地理过程分析与模拟湖北省重点实验室,湖北武汉430079 [2]华中师范大学城市与环境科学学院,湖北武汉430079 [3]武汉大学资源与环境科学学院,湖北武汉430079

出　　处：《测绘学报》2022年第9期1969-1976,共8页Acta Geodaetica et Cartographica Sinica

基　　金：国家自然科学基金(42071455,42071450)。

摘　　要：形状识别是地图空间认知的重要内容之一,结合有效的形状特征向量提取方法和空间认知试验的神经网络方法是提高形状识别的有效途径。本文构建了一种融合了圆形度、偏心率和矩形度等宏观形状特征参量的傅里叶形状描述子作为形状特征向量的神经网络建筑多边形状识别器。首先,利用傅里叶变换和计算几何方法分别提取建筑多边形的傅里叶形状描述子及圆形度、偏心率、矩形度参量,并组成形状特征向量。然后,通过样本数据的训练实现了建筑多边形与形状模板之间匹配的神经网络识别器。结果表明,本文方法相较于以往的方法大幅度提高了精度(达到98.7%),而且特征提取算法不受多边形点数不一致的限制。通过对武汉、郑州两大城市的真实建筑物数据进行形状识别,证实该方法具有较好的识别效果。Shape recognition is one of the important contents of map spatial cognition,and neural network combined with spatial cognitive experiment and its effective shape feature vector extraction are effective ways to improve shape recognition.In this paper,a neural network building-polygon shape recognizer is constructed,which integrates the Fourier descriptors of macro shape parameters such as roundness,eccentricity and rectangularity as shape feature vectors.Firstly,the Fourier shape descriptors,circularity,eccentricity and rectangularity parameters of building polygons are extracted by Fourier transform and computational geometry methods,and the shape feature vectors are formed.Then,the neural network recognizer matching between building polygon and shape template is realized through the training of sample data.The results show that this method greatly improves the accuracy(98.7%)compared with the previous methods,and the feature extraction algorithm is not limited by the inconsistency of polygon points.The shape recognition of real building data in Wuhan and Zhengzhou is carried out,and its information entropy is calculated.This method has good recognition effect.

关 键 词：傅里叶形状描述子 神经网络 模板匹配 建筑多边形形状识别 

分 类 号：P208[天文地球—地图制图学与地理信息工程]