Evaluating Marine Radar Object Detection System Using Yolo-Based Deep Learning Algorithm

Abstract

The purpose of this study is to investigate the application of deep learning techniques, specifically the YOLO (You Only Look Once) object detection model, to enhance the accuracy and efficiency of real-time object detection systems. The study addresses key challenges associated with traditional object detection methods, including issues related to processing speed, scalability, and environmental adaptability. YOLO, being an end-to-end deep learning architecture, offers significant improvements in terms of speed and precision by detecting multiple objects in a single pass through the network, thus enabling real-time detection for autonomous systems. To achieve this, we implement the YOLO framework within a MATLAB environment, training it on diverse datasets to recognize various object classes under different conditions. The procedure involves using labeled training data to train the model, tuning hyper parameters for optimal performance, and evaluating the system’s performance in terms of precision, recall, and average loss during training. The quantitative results of the study demonstrate the model’s ability to achieve high detection accuracy with minimal latency, with the average loss dropping from 0.25 to 0.02 over 50 training iterations with a 93% precision and 90% accuracy. Additionally, the model’s precision, recall, and F1-score consistently outperform baseline methods, further confirming the effectiveness of YOLO in real-time object detection. The study also explores the integration of YOLO with other advanced techniques such as Kalman filters for trajectory prediction and sensor fusion methods, aiming to create a more robust and reliable system for dynamic environments. By demonstrating the application of YOLO in autonomous systems, this research contributes valuable insights into the adoption of deep learning techniques for efficient and scalable object detection. The findings highlight the necessity for continued research into optimizing deep learning models for complex real-time applications, ensuring robust performance across diverse and unpredictable conditions. Furthermore, the study emphasizes the importance of policy development for managing sensor integration, data processing, and computational resources to maximize system efficiency and reliability.