FSD-Net: Underwater Object Detection Model Performance Results

FSD-Net achieves 85.7% AP50 and 82.5% F1-score on the UTDAC2020 dataset and 98.1% AP50 and 97.0% F1-score on the Brackish dataset. The model, proposed by Chao Zhang, uses frequency and spatial domain feature enhancement to improve detection in complex underwater conditions. The dataset, last updated on 2026-04-17, contains experimental results and ablation studies for this novel detection model.

Use Cases

• Benchmarking underwater object detection algorithms based on the reported AP50 and F1-scores.
• Studying the impact of frequency-domain feature enhancement on model performance as described in the methodology.
• Analyzing the effectiveness of multi-dimensional semantic fusion for reducing false detections in degraded visual environments.
• Comparing model performance across different underwater datasets like UTDAC2020 and Brackish.

Strengths

• Performance metrics are explicitly provided, including 85.7% AP50 on UTDAC2020 and 98.1% AP50 on Brackish.
• The description details a clear methodology involving two core enhancement modules (Frequency Attention Convolution and Multi-dimensional Feature Enhancement).
• Results include ablation studies validating the effectiveness of the proposed modules.

Limitations

• The primary data format is a 4.8 MB PDF; the underlying image datasets (UTDAC2020, Brackish) are not directly included.
• Column-level documentation for any potential tabular results is absent; field semantics must be inferred from the paper.
• Row count for any structured performance data is unknown, which may limit suitability assessment.

Provenance

Source

Chao Zhang via figshare.

Collection Method

Likely contains experimental results and model performance metrics from training and testing the FSD-Net model.

Freshness

Last updated 2026-04-17 05:20:56; freshness should be verified.