MSF-DETR: A small target detection algorithm for sonar images based on spatial-frequency domain collaborative feature fusion
View abstract on PubMed
Summary
This summary is machine-generated.This study introduces the Multi-Scale Spatial-Frequency Collaborative Detection Transformer (MSF-DETR) for enhanced small target detection in side-scan sonar images. The novel algorithm improves accuracy and reduces computational load in marine exploration.
Area Of Science
- Marine exploration and engineering
- Underwater acoustics
- Computer vision
Background
- Side-scan sonar imaging is crucial for underwater target detection.
- Small target detection in sonar images presents challenges like limited feature utilization and high computational complexity.
- Existing methods struggle with multi-scale feature fusion and efficient processing.
Purpose Of The Study
- To develop an advanced algorithm for accurate and efficient small target detection in side-scan sonar images.
- To address limitations in frequency domain feature utilization, multi-scale fusion, and computational complexity.
- To introduce the Multi-Scale Spatial-Frequency Collaborative Detection Transformer (MSF-DETR).
Main Methods
- Developed a novel end-to-end algorithm, MSF-DETR, for small target detection.
- Integrated a Multi-domain Adaptive Spatial-frequency Network (MASNet) backbone for dual-domain feature extraction.
- Employed a Hierarchical Multi-scale Adaptive Feature Pyramid Network and an Efficient Sparse Attention Transformer Encoder.
Main Results
- MSF-DETR achieved 78.5% mAP50 and 38.5% mAP50-95 on the SSST-3K dataset.
- Demonstrated significant improvements over the baseline RT-DETR (2.8% and 3.3% respectively).
- Reduced computational complexity by 12.0% and achieved a 71.2 FPS inference speed.
Conclusions
- MSF-DETR offers an effective solution for small target detection in challenging marine environments.
- The algorithm significantly advances underwater sonar image processing capabilities.
- The study highlights the potential of collaborative spatial-frequency processing for sonar image analysis.

