摘要
由于声呐图像受噪声污染严重,导致水下多目标分割存在精度低的问题.为此,提出一种自调整谱聚类结合熵权法进行多特征赋权的水下多目标分割技术.该技术首先通过自调整谱聚类对声呐图像的像素点进行聚类处理,使图像划分为多个独立的区域,然后根据多特征的互补性和冗余性统计每个区域的信息熵、亮度、对比度和狭长度等特征,利用熵权法对多特征进行赋权并筛选出最优的一个目标区域,再将该最优目标区域和所有区域进行多特征相似度匹配,最后根据相似度的匹配结果使用自适应阈值迭代法自动分割出所有的目标区域.实验结果表明没有对噪声干扰区域误分割,分割出的目标区域精度更高,验证了所提方法的有效性.
Sonar image is seriously polluted by noise,which leads to the problem of low precision in underwater multi-target segmentation.Therefore,this paper proposes an underwater multi-object segmentation technique based on self-adjusting spectrum clustering,combined with the entropy weight method.The technology firstly clusters through self-tuning spectral clustering of sonar image pixel clustering processing,so that the image is divided into multiple independent areas.According to the complementarity and redundancy of multiple features,the information entropy,brightness,contrast and narrow length of each region are calculated.The entropy weight method is used to weight multiple features and select the optimal target region.Then,the optimal target region is matched with all regions by multi-feature similarity.Finally,all target regions are segmented automatically by the adaptive threshold iterative method according to the matching results of similarity.Experimental results show that there is no oversegmented of noise interference regions,and target regions segmented have higher accuracy,which verifies the effectiveness of the proposed method.
作者
刘光宇
曹禹
曾志勇
赵恩铭
邢传玺
LIU Guangyu;CAO Yu;ZENG Zhiyong;ZHAO Enming;XING Chuanxi(School of Engineering,Dali University,Dali 671003,China;College of Physics and Optoelectronic Engineering,Harbin Engineering University,Harbin 150001,China;School of Electrical and Information Technology,Yunnan Minzu University,Kunming 650031,China)
出处
《湖南大学学报（自然科学版）》
EI
CAS
CSCD
北大核心
2022年第10期51-60,共10页
Journal of Hunan University:Natural Sciences
基金
国家自然科学基金项目(61761048)
云南省地方本科高校基础研究联合专项资金项目[2019FH001(-066)]
黑龙江省自然科学基金项目(LC2018026)。
关键词
目标
图像分割
聚类
特征选择
熵权法
objective
image segmentation
clustering
feature selection
entropy method
