Surveillance in Maritime Scenario Using Deep Learning and Swarm Intelligence

Abstract

Unmanned Aerial Vehicles (UAVs) have played a crucial role in supporting Search and Rescue (SAR) Operations due to their fast movement capabilities and flexibility. During a search and rescue operation scenario, the time constraint is a crucial parameter, so the required time to detect humans in distress with precision is also a vital part. Modern Deep learning algorithms like CNN also aid in these missions. However, most models and datasets available focus on search and rescue missions on the ground or land. UAV-based search and rescue operations in the Maritime Scenario remain a challenge. This study focused on using deep learning algorithms such as CNN to precisely detect a human in peril with a swarm of drones. At the same time, we emphasize using swarm intelligence algorithms such as Particle Swarm Algorithm (PSO) to effectively find a victim in the shortest time by exploring a massive area. The distinctiveness of this system is that it combines the model with the best accuracy to detect and the best swarm intelligence algorithm for finding targets in the quickest time possible, thus enhancing the surveillance mission. In this research, among VGG16, ResNet50V2, InceptionV3, Xception, and MobileNetV2 models, VGG16 produced an IoU (Intersection over Union) score of 0.62 with a Class Label accuracy of 99.15% and Bounding Box accuracy of 88.74% in CNN part. Along with that, among three different swarm intelligence algorithms, according to the simulation, Particle Swarm Optimization Algorithm took the minimum average time, which is 20.4 units, whereas the Grey Wolf Optimization algorithm and Bat Optimization Algorithm, respectively, took 65.6 and 73.8 units of time.