Deep Learning for Underwater Target Detection and Tracking

This paper presents a method based solely on the YOLO version ۸ architecture to enhance the detection and tracking of targets in underwater environments characterized by various noise and complex conditions. Initially, preprocessing of sonar data and underwater images including spectral normalization, data augmentation through random rotation and controlled noise addition, and removal of environmental noise using Gabor and adaptive median filters is conducted. The YOLOV۸ architecture with integrated compressed inverted blocks is implemented for complex feature extraction, enabling the model to focus on important multi-scale patterns within noisy underwater scenes. Hyper parameters such as learning rate and layer count are finely tuned to optimize detection performance. Results demonstrate that the preprocessing steps significantly reduce noise and enhance image clarity, enabling the proposed model to accurately detect and track small and hidden targets in complex underwater settings. This approach improves detection accuracy, accelerates processing speed, and reduces memory requirements and operational costs, presenting a practical and efficient solution for real-world underwater target detection applications.

کلیدواژه ها:

Deep learning ، Underwater target detection ، Image preprocessing ، estimation