基于GAN-SVC的水下障碍物轮廓构建研究  

Research on contour construction of underwater obstacle based on GAN-SVC

作　　者：唐会林 宋甘琳 周佳加 武杨 TANG Huiin;SONG Ganin;ZHOU Jiajia;WU Yang(Unit of 92213,Zhanjiang 524064,China;College of Intelligent Systems Science and Engineering,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]中国人民解放军92213部队,广东湛江524064 [2]哈尔滨工程大学智能科学与工程学院,黑龙江哈尔滨150001

出　　处：《传感器与微系统》2024年第12期12-15,共4页Transducer and Microsystem Technologies

基　　金：国家自然科学基金资助项目(51609048,51909044,52071108)。

摘　　要：针对由于复杂噪声使得水下无人航行器(UUV)声呐探测数据可靠性下降进而导致障碍物轮廓构建失准的问题,本文提出了一种基于生成对抗网络(GAN)和支持向量聚类(SVC)的水下障碍物轮廓构建算法。为区分复杂噪声点与障碍物点,该算法基于SVC对声呐数据异常点进行初步筛选。针对SVC受参数影响可能导致较小簇误判的问题,利用GAN精确筛选异常点;并对精确的障碍物点进行聚类得到各个障碍物的最优轮廓。通过对湖中障碍物探测数据的轮廓构建仿真验证试验,相比SVC算法,使用本文所提GAN-SVC算法在对2个障碍物进行轮廓构建时,准确度分别提高了79.80%和48.13%。Aiming at the problem that the reliability of underwater unmanned vehicle(UUV)sonar detection data decreases due to complex noise,which leads to the inaccuracy of obstacle contour construction,an underwater obstacle contour construction algorithm based on generative adversarial network(GAN)and support vector clustering(SVC)is proposed.In order to distinguish complex noise points from obstacle points,the algorithm preliminarily screens outliers in sonar data based on SVC.Aiming at the problem that SVC may be affected by parameters and may cause misjudgment of small clusters,GAN is used to accurately screen outliers;and accurate obstacle points are clustered to obtain the optimal contour of each obstacle.Through constructing simulation verification experiments on contour of obstacle detection data in lake,the results show that compared with the accuracy of the proposed GAN-SVC algorithm is 79.80% and 48.13% higher than that of the SVC algorithm.

关 键 词：生成对抗网络 支持向量聚类 异常点检测 轮廓构建 

分 类 号：TP212[自动化与计算机技术—检测技术与自动化装置] TP391.4[自动化与计算机技术—控制科学与工程]