A new technique is presented to detect targets under sea background. Focused on interested areas, the spatial and frequency distributions are first studied, then analytic expression of targets is set up by least square method (LSM). According to the shape, frequency, and analytic expression of targets, a matched biorthogonal wavelet is constructed to better enhance targets and eliminate noises. Results show the usefulness of this method for single frame detection (signal to noise ratio SNR>=1.25), which provides a better performance than classic wavelets and morphological filtering.