JoVE - Journal of Visualized Experiments
JoVE Visualize
Contact Us

ES-YOLO: Multi-Scale Port Ship Detection Combined with Attention Mechanism in Complex Scenes

  • 0School of Remote Sensing and Information Engineering, North China Institute of Aerospace Engineering, Langfang 065000, China.
Sensors (basel, Switzerland) +

|

December 31, 2025

|

December 31, 2025

View abstract on PubMed

Summary

This summary is machine-generated.

The ES-YOLO framework enhances remote sensing ship detection in complex environments using a novel edge perception channel and lightweight modules. This advanced deep learning approach improves accuracy and efficiency for identifying ships in challenging conditions.

Area Of Science

  • Computer Vision
  • Remote Sensing Technology
  • Deep Learning

Background

  • Single-stage algorithms show promise for optical imagery ship detection.
  • Existing methods struggle with complex environments like cloud cover, waves, and dense ship aggregation.
  • Limitations include fixed viewing angles and uniform backgrounds in current models.

Purpose Of The Study

  • To propose the ES-YOLO framework to overcome limitations in complex-environment ship detection.
  • To enhance feature extraction and detail capture using a novel Edge Perception Channel, Spatial Attention Mechanism (EACSA).
  • To reduce computational complexity with a lightweight spatial-channel decoupled down-sampling module (LSCD) and introduce a hierarchical scale structure.

Main Methods

  • Developed the ES-YOLO framework incorporating EACSA, LSCD, and a hierarchical scale structure.
  • Constructed the TJShip dataset using Gaofen-2 imagery, featuring multi-scale targets.
  • Conducted ablation and comparative experiments using the TJShip dataset and various benchmark algorithms.

Main Results

  • ES-YOLO demonstrated improved mean Average Precision (mAP) by 0.83% (EACSA), 0.54% (LSCD), and 1.06% (multi-scale structure) over the baseline.
  • The model achieved superior performance in precision, recall, and F1-score.
  • ES-YOLO outperformed Faster R-CNN, RetinaNet, YOLOv5, YOLOv7, and YOLOv8 in mAP by significant margins (up to 46.87%).

Conclusions

  • The ES-YOLO framework effectively addresses challenges in complex remote sensing ship detection.
  • The integration of EACSA, LSCD, and hierarchical scale structures significantly boosts detection performance.
  • This research offers valuable insights and a robust model for advanced ship detection applications.

Keywords: