Recognition and Classification of Ship Images Based on SMS-PCNN Model

Area of Science:

• Computer Vision
• Artificial Intelligence
• Machine Learning

Background:

• Traditional ship image recognition algorithms struggle with scale-specific features.
• Distinguishing ship categories requires analyzing both coarse-grain hull structures and fine-grain equipment details.

Purpose of the Study:

• To propose a novel Multi-Scale Paralleling CNN (SMS-PCNN) model for enhanced ship image recognition and classification.
• To effectively extract ship features at multiple scales, addressing limitations of existing methods.

Main Methods:

• The SMS-PCNN model employs parallel convolutional branches with varying receptive fields to capture multi-scale features.
• Channel adjustments are implemented to refine feature extraction and reduce redundancy.
• Residual connections are utilized to deepen the network and prevent gradient disappearance.

Main Results:

• The SMS-PCNN model achieved an accuracy of 84.79% on an experimental dataset of open-source ship images.
• Performance tests demonstrated superior results compared to four existing state-of-the-art approaches.
• Ablation experiments confirmed the efficacy of the model's optimization techniques.

Conclusions:

• The SMS-PCNN model offers a significant advancement in ship image recognition and classification.
• The proposed method effectively handles multi-scale feature extraction, leading to improved accuracy.
• The model's architecture and optimization strategies provide a robust solution for complex ship identification tasks.