Weak signal detection has always been a hot spot in the signal processing field. In this paper, the chaotic and large data size characteristics of sea clutter are analyzed, the advantage of the Long and Short Term Memory network (LSTM) is taken to design a weak signal detection method based on deep learning. The reconstructed phase space signal is used as the input of LSTM network, the length of training data is determined by embedding dimension and delay time, and a chaotic prediction model is established to detect weak signals from the prediction error. In order to improve the detection performance, reduce the missing rate of deep learning method for small feature signal, frequency domain conversion of the prediction error is conducted, the spectrum of the prediction error of different distance gates is compared to locate the coordinates of the weak signal. The experimental results show that the sea clutter detection method based on LSTM prediction error frequency domain conversion has strong applicability and higher accuracy, and the detection performance is improved by 30%.

弱信号检测一直是信号处理领域的热点。本文分析了海杂波的混沌和大数据量特征， 借鉴长短期记忆网络（LSTM）的特点，设计了一种基于深度学习的弱信号检测方法。将重构后的相空间信号作为LSTM网络的输入，通过嵌入维数和延迟时间确定训练数据的长度，建立混沌预测模型，从预测误差中检测微弱信号。为了提高检测性能，降低深度学习方法对小特征信号的缺失率，对预测误差进行频域转换，比较不同距离门的预测误差谱，定位弱点坐标。信号。实验结果表明，基于LSTM预测误差频域转换的海杂波检测方法适用性强，精度更高，