Detection of Tiny Objects for Maritime Search and Rescue Operations

Today, deep learning has found extensive usage in a wide number of applications, including healthcare, object detection, smart agriculture, personalized marketing, facial recognition, etc., to name a few. Deep learning has come a long way in the past few decades. In this work, we propose a YOLOv8 inspired approach for the detection of tiny objects in marine search and rescue operations. Experiments have been performed on a dataset of aerial drone floating objects (AFO). The dataset consists of 3647 aerial drone images captured from aerial drone videos that were captured from 35 different places across six countries. The proposed model achieved a Mean Average Precision of 0.973 on the AFO dataset, outperforming previous progressive works carried out on this dataset. This work can prove to be very helpful in real-life scenarios for the detection of humans and other objects trapped in adverse situations.