Abstract: For enhancement and detection of unknown multi-band combined signal under low signal-to-noise ratio, a multi-band combined signal enhancement method based on filter bank design is proposed. Firstly, according to the time-frequency decomposition principle of cochlear, a Gammatone filter bank is designed to divide the data collected by sensor into different frequency bands. Then, according to the differences of intensity and stability between signal and background noise, the normalized variance processing of the data in each frequency band is carried out to establish the decision statistics, by which, the data in different frequency bands are weighted, so that the multi-band combined signal can be enhanced without searching single filter band. The numerical simulation results show that, under the simulation conditions, the signal-to-noise ratio of the unknown multi-band combined signal is increased by more than 6 dB. When the signal frequency band is unknown, this method can enhance the multi-band combined signal and improve the detection effect of the energy accumulation method on unknown multi-band combined signal.