融合时频域特征的舰船识别方法及实验研究

Theoretical and experimental study of ship recognition by fusing feature in time-frequency domains

文章提出了一种融合舰船辐射噪声时频域特征的识别方法,将舰船辐射噪声的线谱特征和线性预测倒谱特征作为输入,分别利用反向传播(Back Propagation, BP)神经网络进行训练、降维及初步判别,并采用加权投票方式,引入置信度算法和拒判机制实现决策级融合识别。实验结果表明,对比基于舰船单一特征的识别方法,利用舰船辐射噪声时频域特征的互补性进行融合识别,减小了单一识别方法误判对总识别率的影响,具有较强的鲁棒性,可有效提高对目标的识别率。

In this paper, a recognition method integrating the time-frequency domain features of ship radiated noise is proposed. By taking the line spectrum features and linear prediction cepstrum features of ship radiated noise as inputs, the back propagation(BP) neural network is used for training, dimension reduction and preliminary discrimination. The weighted voting method is adopted, and the confidence algorithm and rejection mechanism are introduced to realize decision-level fusion recognition. The experimental results show that compared with the ship single feature based recognition method, the fusion recognition is carried out by using the complementarity of the time-frequency domain features of the ship radiated noise, which reduces the effect of the misjudgment of the single recognition method on the total recognition rate, has strong robustness, and can effectively improve the target recognition rate.