FCM-CV水平集算法在沉底小目标声呐图像分割中的应用  

Application of FCM-CV Level Set Algorithm in Sonar Image Segmentation of Small Sinking Target

作　　者：韩婷婷 王璐瑶 周天[1,2,3] 徐超 张丽红[1,2,3] 李海森[1,2,3] HAN Ting-ting;WANG Lu-yao;ZHOU Tian;XU Chao;ZHANG Li-hong;LI Hai-sen(Acoustic Science and Technology Laboratory,Harbin Engineering University,Harbin 150001,China;Key La-boratory of Marine Information Acquisition and Security(Harbin Engineering University),Harbin 150001,China;College of Underwater Acoustic Engineering,Harbin Engineering University,Harbin 150001,China)

机构地区：[1]哈尔滨工程大学水声技术重点试验室,黑龙江哈尔滨150001 [2]工业和信息化部海洋信息获取与安全工信部重点实验室(哈尔滨工程大学),黑龙江哈尔滨150001 [3]哈尔滨工程大学水声工程学院,黑龙江哈尔滨150001

出　　处：《水下无人系统学报》2021年第3期278-285,共8页Journal of Unmanned Undersea Systems

基　　金：国家重点研发计划项目(2018YFF01013401);国家自然科学基金(U1709203,41976176,U1906218);中央高校基本科研业务费专项资金(3072020CFT0501);黑龙江省博士后科研发展基金(LBH-Q18042);黑龙江省自然科学基金项目(ZD 2020D001)。

摘　　要：针对沉底小目标高频声呐图像信混比低,难以与背景分离的问题,提出模糊聚类(FCM)算法与Chan-Vese(CV)水平集算法相结合的分割方法。该方法利用FCM算法得到的隶属度函数自动设定水平集模型曲线的初始演化位置,解决了CV水平集分割因初始位置设置不当而无法得到准确分割结果的不足;同时根据模糊聚类的结果,对水平集演化的控制参数进行估计,使得分割过程更加稳健。通过仿真数据及外场试验数据处理可知,相较于FCM和马尔科夫随机场分割算法,文中算法对斑点噪声不敏感,可分割出完整的边界;相较于常规CV水平集算法,文中算法因粗分割后零水平集的设定更接近目标的边缘,可以在较少迭代次数下即可获得更加精确的分割结果。Aiming at the problem that the signal-to-mix ratio of the small sinking target is low,and it is difficult to separate from the background,this study proposes a segmentation method to use the fuzzy c-means(FCM)algorithm to cooperate with the Chan-Vese(CV)level set.This method automatically sets the initial evolution position of the level set model curve by using the membership function obtained from the FCM algorithm,which solves the problem that the CV-level set segmentation cannot obtain accurate segmentation results because of the incorrect initial position setting.Simultaneously,the control parameters of the level set evolution are estimated according to the results of fuzzy clustering,which makes the segmentation process more robust.Using outfield test and simulation data,compared with the algorithm of FCM segmentation and Markov random field segmentation,the algorithm in this study is not sensitive to speckle noise and cansegmenta complete boundary.Compared with the conventional CV-level set algorithm,the algorithm in this study can obtain more accurate segmentation results with feweriterations.

关 键 词：沉底小目标 模糊聚类算法 CV水平集分割 声呐图像 马尔科夫随机场分割 

分 类 号：TJ630[兵器科学与技术—武器系统与运用工程] TN911.7[电子电信—通信与信息系统]