一种基于修正激活函数的CNN车载毫米波雷达目标检测方法

A Vehicle-mounted Millimeter-wave Radar Target Detection Method Based on Modified Activation Function CNN

为了提高车载毫米波雷达在复杂城市道路环境中目标检测的抗杂波与干扰能力,本文利用卷积神经网络(CNN)特征参数提取和目标分类特性,提出了一种改进的基于CNN的车载毫米波雷达目标检测方法。该方法首先将毫米波雷达回波信号距离-多普勒二维数据运用滑窗进行分割,并采用CNN网络模型处理分割后的二维矩阵,训练二维CNN网络模型及其参数,使其具有提取回波特征并基于特征参数模型进行目标分类的能力,从而实现目标检测功能。通过对卷积神经网络模型结构进行优化,增加批量归一化层,优化Dropout层使得低权重特征失活,自适应地删减部分神经元节点修正该层非线性激活函数,进一步降低了CNN模型目标检测的虚警概率。实验结果表明,在相同虚警概率条件下,CNN网络检测方法目标发现概率优于传统的单元平均恒虚警检测方法,并且在低信噪比的条件下仍然能够保持较高的发现概率;在同等发现概率水平下,修正后CNN网络检测方法的虚警概率较修正前可提高约1个数量级。

In order to improve the anti-clutter and interference ability of vehicle mounted millimeter wave radar in complex urban road environment,this paper uses convolutional neural network(CNN)feature parameter extraction and target classification performance,and proposes an improved method based on CNN-based vehicle millimeter-wave radar target detection method.This method uses sliding window to segment the range Doppler two-dimensional data of millimeter wave radar echo signal,and uses CNN network model to process the segmented two-dimensional matrix to train the two-dimensional CNN network model and its parameters,so that it has the ability to extract the echo features and classify the target based on the feature parameter model.Implement the target detection function.Then,by optimizing the structure of the convolutional neural network model,adding a batch normalization layer,optimizing the Dropout layer to inactivate the low-weight features,and adaptively deleting some neural nodes to modify the nonlinear activation function of this layer,the false alarm probability of the CNN model target detection can be further reduced.The experimental results show that,under the condition of the same false alarm probability,the target discovery probability of the CNN network detection method is better than the traditional unit average constant false alarm detection method,and it can still maintain a high detection probability under the condition of low signal-to-noise ratio;At the same level of discovery probability,the false alarm probability of the modified CNN network detection method can be increased by about 1 order of magnitude compared with that before the modification.