基于视觉传达的舰船航行图像中多目标检测方法  

Multi object detection method in ship navigation images based on visual communication

作　　者：吴军良 毛丽青 姜煜 WU Juniang;MAO Liqing;JIANG Yu(School of Artificial Intelligence,Nanchang Jiaotong Institute,Nanchang 330100,China)

机构地区：[1]南昌交通学院人工智能学院,江西南昌330100

出　　处：《舰船科学技术》2024年第16期170-173,共4页Ship Science and Technology

基　　金：江西省教育厅科学技术研究项目(GJJ2203103)。

摘　　要：针对海洋环境的复杂性以及舰船目标的多样性,研究基于视觉传达的舰船航行图像中多目标检测方法,及时发现潜在的安全隐患。将舰船航行图像从RGB颜色空间转换到CIE Lab模式空间,应用改进HFT(超复数傅里叶变换)模型有效提取舰船多目标显著区域。利用加权处理和Otsu算法划分多层显著区域,并基于先验信息确定舰船多目标候选区域。采用模糊C均值聚类算法对候选区域进行分割,实现舰船多目标的精准检测。实验结果表明,该方法可在云雾覆盖、海洋杂波、船舶尾迹等多种复杂环境下准确检测舰船多目标,具有较高的鲁棒性和实用性。In response to the complexity of the marine environment and the diversity of ship targets,a multi object de-tection method based on visual communication in ship navigation images is studied to timely discover potential safety haz-ards.Convert ship navigation images from RGB color space to CIE Lab mode space,and apply an improved HFT(Hyper-complex Fourier Transform)model to effectively extract multi-target salient regions of ships.Using weighted processing and Otsu algorithm to partition multi-layer salient regions,and determining multi-target candidate regions for ships based on pri-or information.Using fuzzy C-means clustering algorithm to segment candidate regions and achieve accurate detection of multiple targets on ships.The experimental results show that this method can accurately detect multiple targets of ships in various complex environments such as cloud and fog coverage,ocean clutter,and ship wake,and has high robustness and practicality.

关 键 词：视觉传达 舰船图像 多目标检测 显著区域 模糊C均值聚类 候选区域 

分 类 号：TP391.4[自动化与计算机技术—计算机应用技术]