针对高频地波雷达的距离-多普勒图像,提出了一种基于级联分类器的海面目标快速检测算法。首先,根据点目标的形态特征设计了基于点形态检测算子的一级分类器,可以从复杂的杂波背景中快速提取出潜在的目标位置;然后,对于候选目标点,分别提取基于离散余弦变换的纹理特征和基于水平、垂直差分的形态特征,并将两种特征融合为一个31维的特征向量;最后,利用误差自校正极限学习机网络作为第二级分类器,并把每个候选点目标的31维特征向量作为其输入,由此实现对候选目标的精确辨识。实验结果表明:该算法不但提高了目标检测率,而且大大提高了目标检测速度。

A fast sea surface target detection algorithm based on cascade classifier is proposed for range-Doppler images of high-frequency surface wave radar. Firstly, according to the morphological characteristics of point targets, a first-stage classifier based on morphologic point detector is developed to extract potential or candidate target positions from the complex clutter background quickly. Then, for the candidate target points, the texture features based on discrete cosine transform and morphological features based on horizontal and vertical differences are extracted respectively, and then all these features are fused into a 31 dimensional feature vector. Finally,using the 31 dimensional feature vector as its input, an error self-adjustment extreme learning machine is used as the second-stage classifier to further identify the candidate target. The experimental results show that the proposed algorithm can not only effectively improve the target detection rate, but also significantly improve the detection speed.

TN957.51

国家重点研发计划资助项目;海洋公益性行业科研专项资助项目