基于改进SSD的近红外舰船图像目标识别算法  

Target Recognition Algorithm for Near-infrared Ship Image Based on Improved SSD

作　　者：柳碧辉 赵乐源 刘洋 LIU Bihui;ZHAO Leyuan;LIU Yang(Naval Aviation University,Yantai 264001;No.71901 Troops of PLA,Liaocheng 252000)

机构地区：[1]海军航空大学,烟台264001 [2]中国人民解放军71901部队,聊城252000

出　　处：《舰船电子工程》2023年第8期58-63,159,共7页Ship Electronic Engineering

基　　金：国防科技卓越青年人才基金(编号:2017-JCJQ-ZQ-003);泰山学者工程专项经费课题(编号:ts201712072);装备预研共用技术领域基金(编号:JZX7Y20210159100201)资助。

摘　　要：针对舰船可见光图像噪声大、背景冗杂;现有舰船图像目标识别率低、易受到周围海环境的影响等问题,论文提出改进SSD的算法,并与主动轮廓模型相结合对近红外舰船图像进行目标识别。首先将可见光RGB颜色模型转换成HSV颜色模型,然后同时输入到SSD算法与主动轮廓模型分割方法中,获取舰船目标模型的识别框,最后采用非极大值抑制的方法,除去数值差别较大的数据框,最终完成舰船目标图像的识别。经模拟实验验证,论文改进的算法相比原始SSD算法,精确率提高了11.85%,识别时间缩短了一半,证明算法的精确性与有效性。Aiming at the problems of ship visible light image with large noise and redundant background;existing ship image target recognition rate is low,and susceptible to the influence of the surrounding sea environment,this paper proposes an improved SSD algorithm,and combined with the active contour model for near-infrared ships image for target recognition.Firstly,the visible light RGB color model is converted into the HSV color model,and then input into the SSD algorithm and the active contour model segmentation method at the same time.The two algorithms are combined to obtain the recognition frame of the ship target model.Fi⁃nally,the non maximum suppression method is used to remove the data frames with large numerical differences,and complete the recognition of the ship target image.Experimental verification shows that compared with the original SSD algorithm,the improved al⁃gorithm in this paper has an accuracy rate of 11.85%and a half of the recognition time,proving the accuracy and effectiveness of the algorithm.

关 键 词：目标识别 卷积神经网络 红外图像 非极大值抑制 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]