Automated Detection of Vessels in Remote Sensing Images Using Deep Learning Models

This study presents an approach for automated detection
of vessels using convolutional neural networks (CNNs) and
visual transformers (ViTs). The approach involves comparing
different deep learning models including RoboFlow, You Look
Only Once v8 (YOLOv8), RetinaNet, Detection Transformer
(DETR), and Real-Time Detection Transformer (RT-DETR).
The comparison includes a general accuracy metric, mean
Average Precision (mAP), and time performance metrics,
training time, frames per second (FPS), and latency. The
dataset selected for the study is HRSC2016-MS, which
contains 1,680 optical remote sensing images, covering 7,655
labeled instances of vessels and presenting a variety of
conditions, such as lighting, weather, and different scales.
Considering all the metrics, YOLOv8 stands as the best
approach confirming the current popularity of such a family
of object detectors.