Multi Small Object Detection and Prioritized Tracking for Navy Operations using Deep Learning Techniques

Real-time Unmanned Aerial Vehicle (UAV) tracking with deep learning improves situational awareness and surveillance capabilities. There is a significant requirement for prioritized tracking and multi-small item detection in the context of Navy operations. To address this, this paper aims to develop an efficient algorithm for locating and tracking small objects, such as ships and people, in extensive maritime surveillance. It proposes an efficient Multi Small Object Detection and Prioritized Tracking (MSOD-PT) approach. This approach aims to develop a reliable and expandable system that enhances situational awareness and helps naval personnel decide by rigorously prioritizing options in actual Navy operation scenarios. The proposed MSOD-PT adopts YOLOv8 for extremely rapid and accurate object identification, allowing for the simultaneous accurate bounding box localization of several small objects. The allocation of computer resources for significant targets is then optimized through the use of DeepSORT, which is utilized to track using specified criteria and give different IDs to things that have been discovered. The targets are then prioritized according to the user’s monitoring needs. The simulation results, which use the AFO dataset, show that the proposed approach is practical and can react to prioritized targets that are not similar. For the custom YOLOv8 model with a confidence score of 0.6, the accuracy is 0.56 for daytime and 0.33 for Night time, but the precision is 0.8 for daytime and 0.6 for Night time.