Sea-Surface Floating Small Target Detection based on Time-Frequency-Polarization Feature Using BP Neural Network

Due to the complexity of the sea surface environment and the low observability of the target, the detection of sea-surface floating small targets has always been a difficult problem in the field of radar target detection. To tackle this issue, a BP neural network-based target detection method is proposed by full use of time-frequency-polarization (TFP) features. This method consists of three stages. Firstly, the received echoes are transformed into the time-frequency (TF) domain by short-time Fourier transformation (STFT). Then, the TF tri-feature (i.e. kurtosis, skewness, concentration) of targets and sea clutter are calculated from HH, VV, HV, and VH datasets. Finally, the final decision is made by BP neural network to determine whether the cell under the test of radar returns is a target or a clutter. The experiment results show that the proposed method can achieve better performance than the 1-dimensional time or frequency feature-based detector and the support vector machine (SVM) based detector.