基于并行卷积神经网络的地磁方向适配性分析

针对地磁方向适配性分析时人工特征提取主观性较强、所取特征难以表达深层的结构性特征的问题,并为了进一步提高方向适配性分析的准确率,提出了一种基于并行卷积神经网络的地磁方向适配性分析方法。首先,从不同角度建立了地磁场在6个代表方向上的适配性分析图;然后,从同一磁场的不同角度出发,利用卷积神经网络自动完成了特征学习,得到了更为全面的方向适配性特征描述;最后,在并行卷积神经网络所得特征的基础上,利用BP网络建立了地磁方向适配性的分析模型。仿真结果证明,该方法可以有效避免人工特征提取和计算等复杂步骤,实现了地磁方向适配性分析的自动化,而且可以获得优于传统网络和单路卷积神经网络的准确率。

Direction navigability analysis for geomagnetic navigation based on parallel convolutional neural networks

As the manually extracted features are too subjective and can't deeply describe the information of the magnetic area when analyzing the direction matching suitability for geomagnetic navigation,and in order to further improve the analysis accuracy of direction matching suitability,a new analysis method based on parallel convolutional neural networks (CNN) is proposed.Firstly,the matching suitability analysis images of the magnetic areas are established in six typical directions from two different perspectives.Then,the parallel CNN is used to automatically learn and extract the direction matching features in more comprehensive way.Finally,the extracted features of each branch of the parallel CNN are incorporated and input to BP neural network to build the model and analyze the direction matching suitability.Simulation results show the proposed method can avoid human's interference at the most extent and makes it possible to automatically extract direction matching features for magnetic navigation.Meanwhile,the proposed method's accuracy is higher than those of the traditional neural network and the CNN,and it takes only one type of the matching suitability analysis images as its inputs.