A streamlined approach for intelligent ship object detection using EL-YOLO algorithm.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Marine Technology

Background:

• Maritime object detection faces challenges due to low-quality features in complex environments (waves, reflections).
• Existing YOLOv8 detectors struggle with feature understanding and computational efficiency for intelligent ships.
• Unmanned intelligent ships require lightweight, cost-effective object detection hardware.

Purpose of the Study:

• To develop an Efficient Lightweight You Only Look Once (EL-YOLO) algorithm for intelligent ship object detection.
• To improve bounding box regression and feature analysis for maritime objects.
• To create a streamlined and lightweight network architecture for enhanced computational efficiency.

Main Methods:

• Proposed EL-YOLO algorithm based on YOLOv8.
• Incorporated adequate wise IoU (AWIoU) for bounding box regression.
• Utilized shortcut multi-fuse neck (SMFN) for feature analysis and greedy-driven filter pruning (GDFP) for network streamlining.

Main Results:

• EL-YOLO demonstrates significant advancements in detection accuracy and lightweight characteristics.
• Achieved superior performance in intelligent ship object detection using RGB cameras.
• Showcased notable improvements compared to standard YOLOv8 models in maritime scenarios.

Conclusions:

• EL-YOLO offers a promising solution for efficient and accurate object detection in intelligent shipping.
• The algorithm effectively addresses challenges in complex maritime environments.
• EL-YOLO provides a lightweight yet high-performance alternative for maritime object detection applications.