一种改进的SAR图像舰船目标超像素快速检测算法  

An Improved Superpixel Fast Detection Algorithm for Ship Target in SAR Images

作　　者：李雨轩 刘峥[1] 冉磊[1] LI Yuxuan;LIU Zheng;RAN Lei(National Key Laboratory of Radar Signal Processing,Xidian University,Xi’an,Shaanxi 710071,China)

机构地区：[1]西安电子科技大学雷达信号处理全国重点实验室,陕西西安710071

出　　处：《信号处理》2025年第1期113-126,共14页Journal of Signal Processing

基　　金：雷达信号处理全国重点实验室支持计划(JKW202306)。

摘　　要：合成孔径雷达(Synthetic Aperture Radar,SAR)能够不受天气和环境影响,长时间、远距离地高效对目标区域进行探测和数据采集来获取高分辨的观测数据,同时SAR可以多极化、多波段进行观测来获取丰富可靠的信息,以上特性使得SAR在舰船目标检测领域发挥着重要作用。针对传统超像素级恒虚警率检测算法运行效率低的问题,在超像素分割模块提出了一种基于Gamma分布的快速非迭代聚类算法。该算法使用优先级队列来动态实时更新聚类中心,实现了聚类过程的非迭代运算,并且针对SAR图像特点基于Gamma分布改进距离度量,通过衡量图块之间的相似度代替彩色空间距离度量,使得该算法更适合处理灰度图像。针对多目标邻近干扰的问题,本文在CFAR检测模块选用对背景杂波拟合效果好的截断Gamma分布模型。将改进的超像素分割算法与基于截断Gamma分布的检测算法结合,最终设计了一种改进的SAR图像目标超像素快速检测算法。经实验验证,本文提出的改进超像素分割算法在保证分割性能优异的同时,复杂度明显降低;另外截断Gamma分布尽可能地保留了真实的海杂波样本,在统计模型构建上呈现出更高的精确度和更佳的拟合效果,对后续检测算法更有利;在多目标邻近的实测SAR图像中,本文改进的目标超像素快速检测算法具有较高的检测率和执行效率。Synthetic Aperture Radar(SAR)can be independent of the weather and environmental impact,long time,long distance,and efficient detection of the target area and data acquisition to obtain high-resolution observation data.Concurrently,SAR can be multi-polarization,multi-band observation to obtain a wealth of reliable information.The above characteristics make SAR useful for field of target detection on board a ship.The above characteristics make SAR play a crucial role in the field of ship target detection.However,the measured SAR images exhibit multi-target neighborhood interference,and the traditional constant false alarm rate algorithm has poor detection performance and low efficiency.Aiming at the low efficiency of the traditional super-pixel level detection algorithm,a fast non-iterative clustering algorithm based on Gamma distribution is proposed in the super-pixel segmentation module.The algorithm uses a priority queue to dynamically update the clustering center in real time,which achieves the non-iterative operation of the clustering process and improves the distance metric based on the Gamma distribution for the characteristics of SAR images,which makes the algorithm more suitable for dealing with grey scale images.For multi-target neighborhood interference,this study selects the truncated Gamma distribution model with a good fitting effect on background clutter in the CFAR detection module.Combining the improved super-pixel segmentation algorithm with the detection algorithm based on truncated Gamma distribution,an improved super-pixel fast detection algorithm for SAR image targets is finally designed.The algorithm first applies the improved fast non-iterative clustering hyperpixel segmentation algorithm proposed in this paper for SAR images to divide the SAR image into blocks of hyperpixels that contain similar pixels based on the improved distance metric based on the Gamma distribution,which effectively improves the computational efficiency of the segmentation process.Second,an adaptive approach is emplo

关 键 词：合成孔径雷达图像 恒虚警率 超像素分割 舰船检测 

分 类 号：TP751.1[自动化与计算机技术—检测技术与自动化装置]