基于混合神经网络的混沌噪声背景下微弱脉冲信号的检测
Detection of Weak Pulse Signals in the Background of Chaotic Noise Based on Hybrid Neural Networks

微弱信号是指相对背景噪声而言，其信号幅度的绝对值很小、信噪比较低的一类信号，这种信号通常受到各种干扰和噪声的影响，使得其在背景中难以被准确检测或识别。微弱信号可能来自于远距离传输、低功率信号源、弱信号目标等情况。在实际应用中，检测和提取微弱信号是一项重要的技术挑战，因为微弱信号往往包含有用的信息，例如传感器信号、通信信号、生物信号等。有效地检测和分析微弱信号可以帮助我们了解环境、诊断疾病、进行通信传输等。为了提高微弱信号的检测精度，本文构建一种基于注意力机制的CNN-LSTM模型，首先，基于混沌信号对初始值的敏感性及短期可预测性，根据Takens定理对各局部传感器的观测信号进行相空间重构，建立Att-CNN-LSTM模型来对混沌信号进行预测，由此得到单步预测误差，此时，对观测信号的检测问题就可以转化为对一步预测误差的信号检测问题。最后，使用Z检验的方法对微弱信号进行检测，得到局部传感器的检测结果。实验结果表明，本文提出的模型有相较于其他模型有更好的表现。
Weak signals refer to a type of signals with very small absolute values of signal amplitude and low signal-to-noise ratio relative to background noise. These signals are usually influenced by various interferences and noise, making it difficult to accurately detect or identify them in the background. Weak signals may come from long-distance transmission, low-power signal sources, weak signal targets, and other situations. In practical applications, detecting and extracting weak signals is an important technical challenge because weak signals often contain useful information, such as sensor signals, communication signals, biological signals, etc. Effectively detecting and analyzing weak signals can help us understand the environment, diagnose diseases, and facilitate communication transmission. To improve the detection accuracy of weak signals, this paper constructs a CNN-LSTM model based on the attention mechanism. Firstly, considering the sensitivity of initial values and short-term predictability of chaotic signals, the observation signals of each local sensor are reconstructed in phase space according to the Takens theorem. An Att-CNN-LSTM model is then established to predict chaotic signals, obtaining single-step prediction errors. At this point, the detection problem of observation signals can be transformed into a signal detection problem of one-step prediction error. Finally, the weak signals are detected using the Z-test method, and the detection results of local sensors are obtained. Experimental results show that the proposed model in this paper outperforms other models.