大数据驱动的红外移动目标检测

Large data-driven infrared moving target detection

为了精准检测移动目标,提出一种大数据驱动的红外移动目标检测方法。采取空间滤波法预处理红外图像,抑制红外图像背景、增强图像内移动目标边缘,采用Seletive Search策略,通过区域划分算法划分预处理后红外图像为若干块小区域,提取若干个红外移动目标候选区域;为避免相邻红外移动目标候选区域图像间存在帧间差异及虚警,影响移动目标中心位置检测效果,提取移动目标候选区域的灰度特征,并结合运动特征建立加权融合特征,精准定位移动目标候选区域,将移动目标候选区域输入卷积神经网络,网络输出结果即为检测到红外移动目标,利用损失函数判定该目标是否为真实移动目标。实验研究表明:所提方法能够精准定位红外移动目标候选区域,检测出红外移动目标,检测性能较好,拥有较强的收敛性。

In order to accurately detect moving targets,a big data-driven infrared moving target detection method is proposed.The spatial filtering method is used to preprocess the infrared image,suppress the background of the infrared image and enhance the edge of the moving target in the image.Using the seletive search strategy,the preprocessed infrared image is divided into several small regions through the region division algorithm,and several infrared moving target candidate regions are extracted;In order to avoid inter frame differences and false alarms between adjacent infrared moving target candidate area images,which affect the detection effect of the central position of the moving target,the gray features of the moving target candidate area are extracted,and the weighted fusion features are established in combination with the motion features to accurately locate the moving target candidate area.The moving target candidate area is input into the convolution neural network,and the network output result is the detection of the infrared moving target,The loss function is used to determine whether the target is a real moving target.Experimental results show that the proposed method can accurately locate the candidate area of infrared moving target and detect infrared moving target.