共性显著特征引导的SAR图像交通感兴趣区域检测  

Region of Interest Detection for SAR Image Based on Common Salient Feature

作　　者：张文彬 刘芳[2] 孟春雷 闫栋 魏智健 ZHANG Wen-bin;LIU Fang;MENG Chun-lei;YAN Dong;WEI Zhi-jian(Guangzhou North Second Ring Transportation Technology Co.,Ltd.,Guangzhou,Guangdong 510000,China;Academy of Transportation Science,Beijing 100029,China;Beijing GOTEC ITS Technology Co.,Ltd.,Beijing 100088,China)

机构地区：[1]广州市北二环交通科技有限公司,广东广州510000 [2]交通运输部科学研究院,北京100029 [3]北京中交国通智能交通系统技术有限公司,北京100088

出　　处：《公路交通科技》2024年第10期27-36,共10页Journal of Highway and Transportation Research and Development

摘　　要：由于独特的成像机制,SAR图像目标检测存在缺乏光谱信息、斑点噪声干扰和方位角敏感等问题。在光学图像中使用的显著性目标检测方法,大多数不适合直接应用在SAR图像的目标检测中。针对上述问题,从视觉显著性角度入手,提出一种针对SAR图像的联合多图共性显著性分析和单图显著性分析的交通感兴趣区域检测算法。首先提出一种增强的方向平滑滤波器,有效降低SAR图像的斑点噪声。然后,针对多幅图像,提出一个共性显著分析模型,通过提取曲线、纹理和亮度特征等共性特征再聚类,得到共性显著图;针对单幅图像,通过共生直方图生成单图显著图。最后通过一种新颖的融合方法实现最终的多幅图像显著性检测。试验在一个数据集上与4种显著性分析模型进行了比较,模型的ROC曲线位于最上方,曲线下面积为0.909 5,综合评价指标F-Measure值为0.674 45,能准确识别显著区域,抑制背景信息,不会产生误判;与3种经典有效的SAR图像目标检测方法进行了比较,此算法在获得完整和准确的感兴趣区域描述方面表现出色,在比较方法中表现出最高的定位精度。显著性目标检测模型有效利用SAR图像的特征信息并抑制干扰信息,使SAR图像交通感兴趣区域检测和车辆检测结果更准确。Due to the unique imaging mechanism,the SAR image object detection is hindered by issues including lack of spectral information,speckle noise interference and azimuth sensitivity.Most of the salient object detection methods used for optical images are not suitable for SAR images.Addressing the above problems,from the perspective of visual saliency,this study proposed an algorithm for detecting region of interest for SAR images united with the multi-map common saliency analysis and single-map saliency analysis.First,the enhanced directional smoothing filter was proposed to effectively reduce the speckle noise of SAR images.Second,the common saliency analysis model was proposed for multiple images.By extracting common features(e.g.,curves,textures and luminance features,and clustering),the common salient map was obtained.The co-occurrence histogram was employed to generate the single-image saliency map from single images.Finally,a novel fusion method was utilized to realize the final saliency detection of multiple images.Compared with 4 saliency analysis models on a dataset,the ROC curve of the proposed model is at the top;the area under curve is 0.9095;and the comprehensive evaluation index F-Measure value is 0.67445.It can accurately identify the salient areas,suppress the background information,and cannot produce false positives.Compared with other 3 classical object detection methods for SAR images,the proposed algorithm performs well in obtaining the complete and accurate description of region of interest,showing the highest positioning accuracy among the compared methods.The proposed salient object detection model effectively utilizes the feature information of SAR images,and suppresses the interference information,which makes the results of detecting region of interest and vehicle detection more accurate.

关 键 词：智能交通 遥感影像处理 合成孔径雷达图像 共性显著特征分析 联合显著性分析 感兴趣区域检测 

分 类 号：U491[交通运输工程—交通运输规划与管理]