Abstract: In marine monitoring and defense, coastal radars play a vital role in real-time sea surveillance and threat detection. However, radar signals are often corrupted by sea clutter—a complex scattering phenomenon caused by electromagnetic wave interactions with dynamic ocean surfaces. The randomness and time-varying nature of sea clutter degrade target detection performance. To enhance radar accuracy in complex sea conditions, studying sea clutter generation mechanisms and prediction methods is crucial for suppressing interference and improving system reliability.This paper proposes a 1D sea surface target detection algorithm based on phase space reconstruction, requiring only pure sea clutter data for training. By computing time delay and embedding dimension, a CNN-LSTM model predicts short-term sea clutter behavior. Combined with amplitude prediction, the algorithm effectively distinguishes target echoes from clutter. Validation using simulated high-sea-state data and real-world measurements confirms its effectiveness.