SeaLSOD-YOLO: A Lightweight Framework for Maritime Small Object Detection Using YOLOv11

Area of Science:

• Computer Vision
• Artificial Intelligence
• Maritime Surveillance

Background:

• Maritime small object detection is crucial for Unmanned Aerial Vehicle (UAV)-based sea surveillance.
• Challenges include small target size, sea reflections, and wave interference, hindering detection accuracy and real-time performance.

Purpose of the Study:

• To propose SeaLSOD-YOLO, a lightweight algorithm for improved maritime small object detection.
• To enhance accuracy and maintain real-time processing capabilities in complex sea environments.

Main Methods:

• Developed SeaLSOD-YOLO based on YOLOv11, incorporating Shallow Multi-scale Output Reconstruction for detailed feature fusion.
• Integrated SPPF-FD (Spatial Pyramid Pooling with Frequency-Domain adaptive convolution) to enhance high-frequency texture sensitivity and reduce sea-surface noise.
• Employed attention-based feature fusion to prioritize small object features and dynamic multi-scale sampling for optimized feature representation.

Main Results:

• Achieved significant improvements on the SeaDroneSee dataset: precision increased from 75.6% to 81.9%, recall from 62.6% to 73.5%, and mAP@0.5 from 67.9% to 77.0%.
• Demonstrated superior performance in detecting small maritime objects like buoys and lifeboats.
• Maintained a high inference speed of 256 FPS, balancing accuracy with computational efficiency.

Conclusions:

• SeaLSOD-YOLO effectively addresses the challenges of maritime small object detection.
• The algorithm offers a robust solution for real-time UAV-based sea surveillance, balancing accuracy, efficiency, and speed.
• Future research will focus on optimizing attention mechanisms and upsampling for detecting extremely small targets.