Ship Fire Detection Based on an Improved YOLO Algorithm with a Lightweight Convolutional Neural Network Model.

Area of Science:

• Maritime Safety
• Artificial Intelligence
• Computer Vision

Background:

• Ship fires pose significant risks to maritime navigation safety.
• Traditional detection methods lack effectiveness and accuracy due to distance and motion limitations.
• Deep learning offers potential but faces challenges in computational complexity and efficiency for ship fire detection.

Purpose of the Study:

• To develop a lightweight and efficient deep learning algorithm for precise ship fire detection.
• To address the limitations of existing methods in complex marine environments.
• To enhance real-time ship fire warning systems.

Main Methods:

• A modified YOLOv4-tiny algorithm incorporating a multi-scale detection technique.
• Integration of the SE attention mechanism for improved feature extraction.
• Application of image transformation and transfer learning to a small ship fire dataset.

Main Results:

• The proposed I-YOLOv4-tiny + SE model demonstrated superior accuracy and efficiency compared to benchmark algorithms.
• Enhanced detection of small and obscured targets was achieved.
• The model meets real-time ship fire warning requirements in demanding maritime settings.

Conclusions:

• The I-YOLOv4-tiny + SE model offers a precise and efficient solution for ship fire detection.
• The integration of multi-scale detection and SE attention significantly improves performance.
• This approach enhances maritime safety by enabling reliable, real-time fire warnings.