摘要
利用单波段特征进行多类别舰船识别对舰船的图像质量与特征描述算子要求较高,图像质量的下降会直接导致复杂的特征描述算子识别能力下降。为了多舰船目标识别率的提高,针对单波段特征识别能力的不足,设计了目标图像特征,对多特征进行协方差融合,对多波段目标图像,采用特征并行处理,进行协方差特征融合,最后利用融合后的特征对多舰船目标进行识别分类。对比试验结果显示,本文设计的融合特征,在单波段图像上,比直接利用流行的HOG特征识别率高,多波段融合特征的识别率较单波段融合特征更是得到了质的提升,达到了95%以上。
Multi-class ship identification using single band image highly depends on image quality and descriptor of features.Degradation of the image quality will directly influence the identifying capability.In order to increase target recognition rate of multi-class ships,the features of target image were designed,and the features were fused by covari-ance.For multi-band images,parallel processing of the features was used,and multi-ship targets were identified and classified by the fused features after fusion.Experimental results show that recognition rate obtained by using the de-signed fusion features is higher than that obtained by using HOG for single band image.And feature fusion of multi-band images performs better.It can achieve recognition rate of more than 95% .
出处
《激光与红外》
CAS
CSCD
北大核心
2017年第2期239-245,共7页
Laser & Infrared
基金
国家自然科学基金项目(No.61303192)资助
关键词
特征融合
多波段
多类舰船
协方差
目标识别
feature fusion
multi-band
multi-class ships
covariance
target identification