Weak Target Detection based on Deep Neural Network under Sea Clutter Background

To upgrade the performance of the traditional radar target detecting method based on one certain threshold, this paper applies the deep learning network into target detection field, which regards radar target detection as a binary signal classification question. Since sea clutter exhibits non-stationary characteristics with high sea state condition, fractal properties of sea clutter are considered for target detection. In addition, fractal parameters of autoregressive (AR) spectrum are regarded as the feature inputs for deep learning network. Finally, real radar sea clutter data are applied for training the deep learning neutral network, and several datasets are selected to test the detecting performance of the network. From the binary classification results, the proposed method based on deep learning network performs a better detecting performance than traditional CFAR and fractal methods.