| 摘要: |
| 海杂波中弱小目标的检测在军用和民用领域有着重要应用价值。基于径向基函数神经网络的目标检测方法可以检测海杂波中弱小目标，但是训练样本的选择直接影响检测效果。为了减小训练样本对检测效果的影响，提出了基于神经网络集成检测海杂波中弱小目标的方法。根据子网络在验证集上的表现，赋给差异度大的子网络较大的权值，子网络的加权平均得到集成的输出。采用McMaster大学IPIX雷达实测数据的测试结果表明，该方法能够减弱训练样本的选择对目标检测效果的影响，提高检测能力。 |
| 关键词: 海杂波 弱目标检测 神经网络 预测 集成 |
| DOI： |
|
| 基金项目:国家高技术研究发展计划(863计划)项目(2012BAH36B02 ) |
|
| Neural network ensemble approach for detection of weak target in sea clutter |
| LIU Yunfeng,SUO Jidong,LIU Xiaoming,SU Xiaohong |
| () |
| Abstract: |
| It has important values to detect weak targets floating on sea clutter in both military and civilian fields.Weak targets in sea clutter can be detected by radial basis function neural network(RBFNN) prediction error method,but detection results are affected by selection training samples. In order to reduce the impact on target detection by samples,the method based on neural network ensemble is proposed to detect weak targets in sea clutter.According to the subnetwork's performance in validation data set,the subnetwork with large difference will be assigned larger weight value. And output of ensembles is constituted of weighted average of subnetwork′s output. The method can decrease the impact of the training samples selection on target detection effect and enhance the detectability of weak targets embedded in a sea by live recorded sea returns collected by the McMaster IPIX radar. |
| Key words: sea clutter weak targets detection neural network prediction ensemble |
