To address the low efficiency caused by point-by-point analysis and serial processing in massive IQ data, a fast pulse detection method based on parallel downsampling and segmented progressive processing is proposed. In the proposed method, the optimal data allocation for each thread is derived according to disk I/O characteristics, and optimized data partitioning and multithreaded parallel processing are implemented. A three-stage progressive detection strategy is then performed independently in each thread. First, the noise-floor energy is estimated from the instantaneous energy of sampled points, and the detection threshold is determined using the Z-score method. Candidate pulse segments are screened by coarse-grained sampling, pulse edges are identified by fine-grained sampling, and pulse start and end positions are located by high-resolution sampling. Finally, split pulses at segmentation boundaries are identified and merged using a multi-point continuity detection strategy. Experimental results show that the proposed method reduces detection time by 95.7% compared with conventional methods and improves processing efficiency by 43.2% after multithreading.