基于知识表示与推理的遥感图像中的对象识别1
Object Recognition in Remote Sensing Images Based on Knowledge Representation and Inference
完颜丹丹 1孙士保2
作者信息
- 1. 郑州工商学院信息工程学院,郑州 451400
- 2. 河南科技大学软件学院,河南洛阳 471000
- 折叠
摘要
为了实现对遥感图像中对象的识别,提出了一种知识表示与推理的对象识别方法.该方法首先执行图像分割,并对每个分割后的区域提取一组与光谱、空间和背景属性相关的特征来表征,以便进行对象识别;然后以知识形式化的本体方式对一个领域进行建模,并定义了一组概念(如建筑物、植被、道路、水等)以及它们的特征和彼此之间的关系.为了给每个区域分配一个语义含义,提出了一种面向特征的对象和本体概念之间的匹配,并给出了计算本体的原始匹配度量和基于启发式方法的遍历过程,从而实现整个图像的识别.基于高景一号卫星影像图的实验结果表明该方法不仅是有效和鲁棒的,而且在精度、召回率和F-度量的平均值方面优于现有的其他图像识别方法.
Abstract
In order to realize object recognition in remote sensing images,basing on knowledge representation and inference,an object recognition method is proposed in this paper;In proposed method,image segmentation is performed firstly,and each segmented region is characterized by means of a set of features,which are related to spectral,spatial and contextual attributes,are extracted from each segmented regionin in order to perform object recognition;A domain is then modeled in a knowledge-formalized ontology that defines a set of concepts(such as buildings,vegetation,roads,water,etc.),their characteristics and their relationships to each other.In order to allocate a semantic meaning to each region,a feature-oriented matching between an object and the concepts of ontology is proposed,and the original matching measure of the ontology computing and the traversal process based on heuristic are given to realize the recognition of the whole image;The experimental results based on Super-View-1 satellite image show that the proposed method is not only efficient and robust,but also superior to other existing image recognition methods in terms of the average values of precision,recall rate and F-measure.
关键词
图像识别/语义/地理本体/知识形式化/匹配分数/遍历/召回率Key words
image recognition/semantics/geographical ontology/knowledge formalization/matching score/traversal/recall rate引用本文复制引用
出版年
2024
NETL
NSTL
万方数据