Digital bandwidth interleaving (DBI) technology is a method that can simultaneously improve the bandwidth and sampling rate of a sampling system. However, its waveform reconstruction algorithm is very complex, and the realization of real-time waveform reconstruction requires a great number of system parallel computing resources. This paper proposes a frequency-domain waveform reconstruction method for a DBI sampling system to improve the waveform reconstruction efficiency. This method completes digital mixing, filtering, and corrections in the frequency domain, and recovers the waveform from the discrete frequency domain information. This method takes advantage of the computational efficiency advantages of the frequency-domain filtering method compared to the time domain filtering. And it integrates multiple time-domain filter coefficients in the DBI system into a frequency response coefficient, so that one frequency-domain filtering is equivalent to multiple time-domain filtering operations. This paper also introduces a polyphase frequency-domain filtering structure, which has great advantages in computational complexity compared to the classical time-domain parallel polyphase filtering structure. To verify the effeciency of the method, an evaluation platform with 8 GHz bandwidth and 20 GSPS sampling rate is implemented to illustrate its performance. Both the classical time domain method and the proposed frequency domain method are verified in MATLAB and are compared for performance and resource consumption in FPGA. Experimental results show that the proposed method saves 10% look up tables (LUTs), 26% flip-flops(FFs), and 47% digital signal processing (DSP) slices compared to the classical time-domain method. Under the condition that the word length is less than 6 bits, the proposed method improves SFDR by more than 5 dB and SNR by more than 10 dB compared with the classical method.