Abstract: Dynamic channelization technology, a key technology for realizing space-based broadband spectrum sensing, possesses the capability of real-time decomposition and parallel processing of broadband signals, which can alleviate the computational and processing pressure on onboard resources. To address the cross-channel issues in broadband channels, this paper employs a polyphase filter bank with perfect reconstruction characteristics to establish an analysis-synthesis joint processing system, and proposes a joint time-frequency domain cross-channel decision algorithm based on an optimized adaptive threshold constant false alarm detection (Optimized-CFAR), achieving adaptive fusion and accurate reconstruction of cross-channel signals. Simulation results demonstrate that the detection probability reaches 98.6% at a signal-to-noise ratio (SNR) of 15 dB, with the amplitude distortion of the reconstructed signal being approximately 0.0048 dB and the reconstruction fidelity achieving 0.972. The FPGA implementation complexity of the proposed algorithm is reduced by 13.2% compared to existing advanced schemes.