Authors

• Naga Venkata Rishika Guggilam, Department of Computer Science and Engineering, VR Siddhartha Engineering College(A), Siddhartha Academy of Higher Education (Deemed to be University), Vijayawada, India
• Rupa Chiramdasu, Department of Computer Science and Engineering, VR Siddhartha Engineering College(A), Siddhartha Academy of Higher Education (Deemed to be University), Vijayawada, India
• Akhil Babu Nambur, Department of Computer Science and Engineering, VR Siddhartha Engineering College(A), Siddhartha Academy of Higher Education (Deemed to be University), Vijayawada, India
• Naveena Mikkineni, Department of Computer Science and Engineering, VR Siddhartha Engineering College(A), Siddhartha Academy of Higher Education (Deemed to be University), Vijayawada, India
• Yaodong Zhu, Jiaxing University School of Information Science and Engineering, Jiaxing Zhejiang 314001, China
• Thippa Reddy Gadekallu, Division of Research and Development, Lovely Professional University, Phagwara, India.

Journal: Computers & Security, Elsevier

Introduction

The integration of deep learning with security is becoming increasingly vital, especially in the domain of maritime surveillance. This field faces significant challenges, including low detection accuracy, high computational complexity, and inefficient GPU utilization. Additionally, deep learning applications in this domain require extensive datasets to achieve high accuracy. The paper introduces an innovative privacy-preserving deep learning-based vessel monitoring system that not only tracks and classifies vessels but also emphasizes data integrity and authenticity.

Objectives

• Enhance Detection Accuracy: Improve the precision of vessel detection and classification over existing models.
• Ensure Data Security: Implement mechanisms to ensure data integrity and authenticity.
• Optimize Computational Efficiency: Address the challenges of high computational complexity and limited GPU utilization.

Methodology

The proposed system integrates the YOLOv8 (You Only Look Once, version 8) deep learning model with the SHA-256 cryptographic hash function. The integration of YOLOv8, known for its real-time object detection capabilities, is designed to enhance vessel monitoring by ensuring both accuracy and speed. SHA-256 is used to secure the data, ensuring its integrity and authenticity throughout the monitoring process.

Data and Architecture

Dataset: The system utilizes a balanced dataset consisting of 693 photo-realistic video sequences. This dataset is crucial for training the model to accurately detect and classify vessels under various conditions.

Model Components:

• CSPDarkNet53: This is used as the backbone of the network for feature extraction. It enhances the model's ability to capture important features from input images.
• Spatial Pyramid Pooling (SPP): This layer is employed to increase the receptive field and to handle objects at different scales.
• Head Layer: Responsible for the final detection and classification of vessels.

Results

The integration of YOLOv8 with the advanced architectural components and the use of a class-balanced dataset resulted in significant improvements:

• Precision Improvement: The proposed model demonstrated a 9.3% increase in precision compared to YOLOv7. This significant enhancement in precision highlights the system's superior performance in detecting and classifying vessels.
• Data Security: The use of SHA-256 ensures that the data collected and processed by the system remains secure and tamper-proof, addressing concerns of data integrity and authenticity.

Conclusion

The expert system developed in this study provides a robust solution for maritime surveillance by integrating advanced deep learning techniques with cryptographic security measures. By achieving a notable improvement in precision and ensuring data security, this system sets a new standard for vessel detection and monitoring.

Implications and Future Work

The study's findings have significant implications for the fields of maritime security and surveillance. The improved precision and data security make this system highly suitable for practical deployment in real-world scenarios. Future work could explore further optimizations in model architecture, the use of larger and more diverse datasets, and the integration of additional security measures to further enhance the system's capabilities.