一种基于车载探地雷达阵列的地雷实时检测方法

Real-time method for landmine detection using vehicle array GPR

相对金属探雷器和单通道探地雷达,采用车载探地雷达阵列探测系统能对雷场进行快速检测.但是,雷达阵列单位时间内采集的数据量大,且器件间的干扰信号多,使得实时地雷目标检测算法更有难度.本文实现了一套完整的基于车载探地雷达阵列的实时地雷目标检测方法.首先,在车体前进方向上对阵列雷达数据进行预处理,以突显目标响应,压制干扰信号;然后,在阵列排布方向上计算检测窗口内图像的方向梯度直方图(HOG)特征,利用训练好的支持向量机(SVM)预测识别疑似地雷;另外,采用权重判断法和能量分析法剔除部分虚假目标;最后,在车体前进方向上,基于地雷目标尺寸大小压制部分虚警并确定地雷目标,同时对地雷目标进行定位,确定地雷埋深,在深度可调的深度切片上实时显示地雷目标.试验数据表明:该方法能对地雷目标进行实时检测,并明显降低了地雷目标的误检率.采用车载探地雷达阵列探测系统对雷场进行实时地雷目标检测是可行的,但是,仅通过探地雷达技术仍无法同时兼顾高检测率和低虚警率.同时使用多种技术手段,采用多特征融合决策探雷是一种必然趋势.

Ground Penetrating Radar(GPR)is an effective method to detect subsurface target.Compared with metal detector and single-channel GPR,vehicle array GPR can quickly scan minefield.But,the amount of data collected by radar array per unit time is large,and more noises add because of the complex instruments.Data processing will be more difficulty.Here,we present a real-time landmine detection method used on a 3-dimensional(3-D)dataset of vehicle array GPR.Firstly,the preprocessing is operated in each channel(down-track direction)to prominent the abnormality.Then,the Histogram of Oriented Gradient(HOG)method and Support Vector Machine(SVM)are performed to flag potential locations of interest among the separated 2-dimensional(2-D)GPR image in the cross-track direction.The weight judgment method and Energy Distribution method is used to exclude non-mine targets.Thirdly,we confirm targets on the horizontal plane.Finally,the location of the landmine is shown on the time slice which can adjust its time.Experimental results show that the approach we presented can detect the landmine targets in real-time with a low false alarm rate.It means vehicle array GPR can detect landmine in real-time,however,only array GPR can’t simultaneously consider the high detection rate and the low false alarm rate.Multiply a variety of sensors and fusion decision-making will be a trend.