Lightweight deep neural network for navigation safety assessment system detecting small floating objects on sea surface

This research explores the use of deep neural networks for detecting floating objects on the sea surface in a ship’s route. The study focused on assessing the performance of lightweight neural networks, considering factors such as size, speed, and efficiency. A crucial aspect of the evaluation was the accurate classification of sea surface images, determining the safety level associated with detected floating objects like fishing nets, plastic debris, and buoys, while also ensuring proper identification of marine animals. The proposed system is intended to support decision-making by alerting the watch officer or helmsman to potential hazards, thereby reducing the risk of missing critical obstacles. To assess real-time functionality, additional time-based tests were conducted. The findings confirm the feasibility of developing a lightweight detection system using convolutional neural networks to measure real-time safety value of the navigational situation.