CNN-Based Sea-Surface Target Detection Using Continuous Wavelet Transform

Building a high-performance target detector under sea clutter background has always been a difficult but vital problem leading to various detection methods. In this work, we propose a ResNet-CWT detector. ResNet is adopted to gain a higher target detection probability ( \(P_d\) ) and solve the network degradation problem as the network is deepening, while CWT is applied to extract time-frequency feature effectively. The result of tests shows that the \(P_d\) of ResNet-CWT detector is higher than other CNN detectors including LeNet-5-STFT, AlexNet-STFT and ResNet-STFT. Also, when increasing the observation window length, the \(P_d\) of target rises and the false alarm rate decreases.