Abstract: With the rapid proliferation of marine resource exploitation and the continuous expansion of sophisticated underwater observation networks， underwater acoustic communication （UAC） has firmly established itself as the sole effective and viable technical means for medium-to-long-distance information transmission in aquatic environments. Consequently， the requirements for real-time processing capabilities and systemic reliability in UAC are becoming increasingly stringent. Within the architecture of UAC systems， preamble signal detection serves as a critical and foundational stage. The performance of this stage dictates the overall efficacy of the communication link： a low detection rate fundamentally constrains communication efficiency and throughput， whereas a high false alarm rate indiscriminately causes unnecessary activation of communication equipment. This frequent， unnecessary awakening severely exacerbates system power consumption， which is particularly detrimental for battery-constrained underwater nodes. However， the intrinsic physical properties of the underwater acoustic channel present formidable obstacles. The channel is notoriously characterized by severe multipath effects， profound Doppler frequency shifts， and highly complex， dynamic ocean ambient noise. These environmental factors pose significant challenges to the design and implementation of preamble signal detection algorithms that must simultaneously achieve high robustness and low power consumption. Traditionally， the normalized matched filter （NMF） algorithm has been adopted extensively due to its implementation simplicity. Nevertheless， in scenarios afflicted by severe channel multipath interference， the correlation peaks generated by the NMF are highly susceptible to severe dispersion. This dispersion degrades the signal-to-noise ratio at the receiver， thereby severely limiting the overall detection performance. To overcome these traditional limitations， recent studies have focused on reconstruction-based detection methods that exploit the inherent sparsity of the underwater acoustic channel. Among these， the sparse detector based on the matching pursuit （MP） algorithm has demonstrated superior detection performance compared with conventional methodologies. Despite its accuracy， the MP-based approach suffers from prohibitively high computational complexity， rendering it largely impractical for real-time applications in scenarios characterized by dense multipath components. To address the critical trade-off between detection performance and computational efficiency， this paper proposes a novel， rapid preamble detection method based on iterative hard thresholding （IHT） reconstruction. By designing a specific dictionary matrix tailored to the operational mechanics of the IHT algorithm， the proposed method significantly reduces the computational overhead while preserving the high performance of the reconstruction algorithm. Furthermore， based on the sequential reconstruction results obtained via a sliding multi-window mechanism， this study constructs first-order and second-order statistical metrics to formulate a novel cascaded dual-threshold detection strategy. The proposed strategy employs a two-stage verification process. In the second stage， second-order statistics are utilized to perform a secondary， refined discrimination on signal samples that fall within the ambiguity zone of the first-order statistics. This hierarchical approach substantially reduces computational complexity and processing latency while maintaining a detection performance that closely approximates that of the computationally intensive MP sparse detector. To validate the proposed methodology， comprehensive field experiments were conducted in the Qiandao Lake. The experimental results demonstrate that the average processing time is reduced to merely 11.22% of that required by conventional methods. Ultimately， the proposed IHT-based cascaded detection method effectively satisfies the stringent requirements of modern UAC for real-time processing and ultra-low power consumption.