I. Introduction
HF skywave OTH radar can detect targets at ranges of thousands of kilometers taking advantage of the reflection of high frequency (HF) electromagnetic wave by ionosphere [1]. Ship detection is a significant application and has been achieved within long coherent integration times (CITs) in the scenario that the ship's Doppler frequency is not covered by the ocean clutter's [2]–[3]. Within short CITs, i.e. fewer than 10 seconds [4], the short sequence of received data results in poor resolution in Doppler frequency domain by Fast Fourier Transform (FFT) and low signal-to-noise ratio (SNR) after coherent integration, which increases the missed alarm rate. Dealing with this problem, conventional methods can be divided into two sorts. Some employ super-resolution algorithms such as AR model spectrum estimation or MUSIC spectrum estimation to realize recognition of ship targets [5]–[7]. And the others apply accurate ocean clutter suppression methods such as ROOT ocean clutter cancellation, ocean clutter suppression via eigenvalue-decomposition (EVD), singular value decomposition (SVD) and estimation of the subspace from Doppler (ESVID) [8]–[13]. Generally speaking, the methods of super resolution algorithms provide high resolution spectrum, but the amplitudes can't accurately represent the power of the corresponding Doppler frequencies, which means the spectrum can't be the sufficient foundation for ship detection. Methods of Root ocean clutter cancellation, SVD and ESVID only make use of the information of the interested resolution cell which result in that recognition between the interested target and ocean clutter is not convincible without priori knowledge. The EVD method applies the information of the adjacent cells to suppress the ocean clutter of the interested resolution cell, which realizes auto clutter suppression in normal cases. However, the complex environmental conditions such as ocean current can attenuate the correlativity of ocean clutter between the interested resolution cell and the adjacent cells in the same subfield under surveillance, in which case the effectiveness of the EVD method degrades.