A Training-Free Selective-Processing Workflow for In Situ Marine Particle Fields Using Parallel Phase-Shifting Digital Holography

In situ marine particle-field observation by parallel phase-shifting digital holography (PPSDH) produces long image sequences under real deployment conditions, but exhaustive full-frame reconstruction and segmentation are computationally expensive when many frames are low contrast and particle-like targets occupy sparse regions. This paper presents a training-free two-stage selective-processing workflow for a 9521-frame coastal South China Sea PPSDH campaign. Stage 1 uses an amplitude-derived contrast metric as a campaign-specific pruning rule to form a retained-frame pool, and Stage 2 combines coarse reconstruction, candidate filtering, valid-field gating, and ROI merging for ROI-restricted reconstruction and segmentation. Stage 1 retained 6970 frames, corresponding to 73.2% of the full sequence. On a balanced 120-frame benchmark, Stage 2 achieved a spatial-support reduction ratio of 49.9% ± 12.0%, and the complete workflow provided a 5.66-fold end-to-end speedup relative to a matched full-frame baseline. The efficiency gain was accompanied by a measurable fidelity cost, with a baseline-matched correspondence rate of 0.612 and a count-based yield gap of 0.287, mainly associated with small or weak targets within the selected ROI support. These results show that the proposed workflow can support computation-aware review of real marine PPSDH particle fields by efficiently prioritizing informative frames and particle-like regions for downstream visual assessment.