Confidence-Convergence-Based Sequential Recognition of Underwater Moving Target Behaviors under Limited Observations
Description
This dataset provides the figure source data, generated figures, and plotting/audit scripts associated with the manuscript “Confidence-Convergence-Based Sequential Recognition of Underwater Moving Target Behaviors under Limited Observations”. The data support the simulation-based evaluation of a reliability-aware sequential recognition method for underwater moving target behavior recognition under limited observations. The package includes CSV files for overall method comparison, coverage-risk analysis, observation-window effects, confidence-evolution examples, ablation studies, robustness tests under noise, missing observations and delay, and additional sensitivity tests. It also includes the corresponding publication figures and scripts for figure generation and data-quality auditing. The dataset is intended to support reproducibility of the numerical results and visualizations reported in the manuscript.
Files
Steps to reproduce
The data were generated from a simulation-based evaluation workflow developed for the manuscript “Confidence-Convergence-Based Sequential Recognition of Underwater Moving Target Behaviors under Limited Observations”. No physical instruments, biological materials, or reagents were used. Three representative underwater moving target behavior types were simulated as abstract motion-evolution patterns under limited observation windows. Sequential observation features were generated to represent motion-related and sensing-quality information, including speed trend, heading-change variation, bearing-rate variation, motion consistency, detection confidence, and observation quality. Uncertain sensing conditions were introduced by adding observation noise, missing observations, and delayed measurements. The recognition experiments were then conducted using the proposed confidence-convergence-based sequential recognition method and several baseline methods, including static classification, Bayesian sequential recognition without convergence assessment, threshold-only sequential recognition, HMM-based recognition, GRU-based recognition, and abstention-capable baselines. The reported data include overall method comparison, coverage-risk analysis, observation-window sensitivity, representative confidence-evolution cases, ablation studies, robustness tests, and sensitivity analyses. The dataset was organized as CSV source files for each figure and table. The included Python scripts were used to audit the numerical consistency of the source data and regenerate the publication figures. This workflow was designed to make the numerical results and visualizations in the manuscript reproducible from the provided data files.
Institutions
- Wuhan University of TechnologyHubei, Wuhan