特征检测是提高海面小目标检测的有效途径。针对低维特征检测概率低和高维特征虚警难控制的问题,文中提出一种基于虚警可控随机森林的高维特征检测方法。首先,从时域、频域、时频域等多域提取多维特征,将检测问题转换为高维特征空间中的两分类问题;其次,通过仿真含目标回波,获取海杂波和含目标回波的两类均衡训练样本;然后,将随机森林算法引入到高维特征空间中,建立分裂因子和虚警率的函数关系,获得虚警可控的判决区域;最后,基于IPIX 实测数据验证所提检测器具有一定的性能提升,满足实际雷达恒虚警检测的要求。

Feature detection is an effective way to improve the detection of small sea-surface targets. Aiming at the problems of low detection probability of low-dimensional features and difficult control of high-dimensional feature false alarms, a high-dimensional feature detection method based on random forest with controllable false alarms is proposed in this paper. First, multi-dimensional features are extracted from multiple domains of time domain, frequency domain, and time-frequency domain. The detection problem is converted into a two-class classification problem in high-dimensional feature space. Second, two types of balanced training samples including sea clutter and target echo are obtained by simulating returns with target. Third, random forest algorithm is introduced into high-dimensional feature space, and function expression of the splitting factor and the false alarm rate is established to obtain the control region of false alarm. Finally, it is verified by the IPIX measured data that the proposed detector has a certain performance improvement and meets the requirements of real radar with constant false alarm detection.

TN957

国家自然科学青年基金资助项目(61901224);南京信息工程大学人才启动经费资助课题