基于联合分布适配的水下声源测距算法研究

水下声源被动测距基于接收数据中声源辐射的声压信号，通过特定方法在空域中搜索声源位置参数，是一个参数估计问题。对于参数估计问题，机器学习方法通常将其转化为分类问题，相比于传统匹配场处理(MFP)具有更准确的估计能力，并且无需先验的声场环境信息。但当训练数据和测试数据的概率密度函数服从不同的分布或者训练数据严重不足时，传统机器学习方法下的分类器预测效果通常较差。因此，该文提出基于联合分布适配(JDA)的水下声源测距算法，该算法使用JDA寻找恰当的变换矩阵进行数据映射，从而减小不同数据域间分布差异，实现源域到目标域的迁移。对经过JDA后数据进行实验的结果表明，JDA可以有效降低在不同时间和不同方位的水声场中获取航迹数据之间的差异，使得基于源域训练的分类器对目标域预测结果的均方根误差(RMSE)和平均绝对误差(MAE)降低了超过30%，从而实现对声源更准确的距离估计。

Research on Underwater Source Ranging Algorithm Based on Joint Distribution Adaptation

Underwater source passive ranging is based on the pressure radiated by the source in the received data. It is a parameter estimation problem to search for source position parameters in the airspace through the method. Parameter estimation problems are usually converted into classification problems by machine learning methods, which have more accurate estimation capabilities than traditional Matched Field Processing (MFP) and with needless prior sound field information. However, when the probability density of training data and test data follow different distributions or the training data is insufficient, the effect of the classifier under traditional machine learning methods is usually poor. Therefore, an underwater target source ranging algorithm based on Joint Distribution Adaptation (JDA) is proposed to find an appropriate transformation matrix for data mapping, thereby reducing the distribution differences and realizing the migration between source and target. The experimental results indicate that JDA can effectively reduce the differences between the track data obtained in the underwater acoustic field at different times and orientations, thus target could be predicted by classifier based on the source training. The resulting Root Mean Square Error (RMSE) and Mean Absolute Error (MAE) are reduced by more than 30%, enabling more accurate distance estimates.