ID 原文 译文
54487 其次,通过极化域和特征域的多维特征线性融合,将多维特征压缩到3D特征空间中,获得高维度信息的同时减少维度计算代价。 Second, multi-dimensional features are compressed into 3-dimentional feature space by the linear fusion in polarization domain and feature domain, which can obtain high-dimensional information and reduce dimensional computational cost at the same time.
54488 然后,结合凸包学习算法获得3D判决区域,实现异常检测。 Third, convex hull learning algorithm is used to obtain the 3 D decision region and realize the anomaly detection.
54489 最后,基于IPIX实测数据的实验结果表明:相对现有的极化检测器,提出的检测器具有25%以上的显著性能提升。 Finally, experimental results via IPIX data show that the proposed detector can attain significant performance improvement of more than 25%, relative to the existing polarization detectors.
54490 为了提高浅海脉冲噪声环境下水声通信信号调制识别的性能和实用性,提出了基于降噪自编码器和卷积神经网络的调制识别方法。 To improve the performance and practicability of modulation recognition of underwater acoustic communication signals in impulse noise environment of shallow sea, a modulation recognition method based on denoising automatic-encoder(DAE) and convolutional neural network(CNN) is proposed.
54491 算法构造了联合降噪自编码器和卷积神经网络的框架,利用降噪自编码器对含噪声信号进行降噪处理, First, a joint construction of DAE and CNN is proposed, in which the DAE plays an important role to suppress impulsive noise,
54492 利用卷积神经网络对降噪信号的功率谱图进行调制方式的分类识别。 and the CNN is used to classify the power spectrum of signals after noise reduction.
54493 为了解决目标水域水声通信信号训练样本不足的问题, Meanwhile, in order to solve the problem of insufficient training samples of underwater acoustic communication signals in target waters,
54494 采用迁移学习思想,利用典型声剖面构造水声通信信号训练数据集,采用两步迁移策略提升小样本条件下的水声信号调制识别能力。 the idea of transfer learning is adopted. The transfer learning data set is constructed by the typical acoustic profile and the two-step transfer learning strategy is used to improve the modulation recognition ability of underwater acoustic signals with small samples.
54495 仿真实验和实测数据验证了本文方法的有效性。 Simulation results and practical signal tests demonstrate the effectiveness of the proposed method.
54496 与现有算法相比,本文所提方法具有较高的识别率,并且提升了目标信道数据不足条件下的识别性能。 Compared with the existing methods, the proposed method improves the accuracy rate of modulation recognition in impulse noise environment even with insufficient target data.