基于图像分割的地雷目标检测

Landmine detection based on image segmentation

前视地表穿透雷达能够对车前安全距离外的区域成高分辨率图像,利用前视地表穿透雷达进行地雷探测是一个复杂环境下的微小目标检测问题。受非平稳背景干扰,传统检测算法探测性能有限。该文提出一种基于图像分割的背景估计及地雷目标检测方法。连续多帧图像用来估计出耦合信号,为降低计算量,将成像后估计改进为回波域估计后成像。对剔除耦合信号后的图像实施均衡以保证各处增益相同,均衡后图像利用二维Otsu算法分割出能量较强的区域,降低后续检测过程中强目标对杂波统计的影响,最终获得比传统检测算法更好的探测性能。该文同时还提出一种快速算法用于实时系统。通过实测数据验证,该方法可以有效改善前视地表穿透雷达的探测性能。

Forward-Looking Ground Penetrating Radar has the capability of forming two-dimensional high-resolution images of subsurface objects from a standoff distance.Landmine detection using forward-looking ground penetrating radar was a problem of small targets detection.The performance of traditional detection algorithms degraded because of the presence of non-homogenous environment. In this paper,a segmentation-based method was proposed,which was used to estimate the background and detect landmines.Self-signatures were computed by multi frame images.To decrease computing complexity,estimation after imaging was improved to imaging after estimation bv echo data.The images eliminating self-signatures were balanced to ensure the gain of images equal everywhere.The images after balanced were segmented by two-dimension Otsu algorithm.Target areas were masked and decreased the influence to the statistic of clutters.The performance of this method was better than traditional algorithms.A fast algorithm was also proposed to the application of real-time system.It is proved by real data that the method can increase the detection performance of forward-looking ground penetrating radar a lot.