Citation

Mehran A, Tehsin S, Hamza M. Spat. Inf. Res. 2023; 31(1): 61-72.

Copyright

(Copyright © 2023, Korean Spatial Information Society, Publisher Holtzbrinck Springer Nature Publishing Group)

DOI

10.1007/s41324-022-00482-1

PMID

unavailable

Abstract

Detecting ships from satellite images is a challenging task in the domain of remote sensing. It is very important for security, traffic management and to avoid smuggling etc. SAR (Synthetic Aperture Radar) is mostly used technology for Maritime monitoring but now researchers are increasingly studying Optical Satellite Images based technologies. Image processing and Computer Vision techniques were previously used to detect ships. In this work, Convolutional Neural Network based approach is used to detect ships from the satellite imagery. Several Deep Learning models have been used and tested for this kind of task. We used state of art model Inception-Resnet that is pre trained on Image-Net dataset. We used the dataset "Ships in Satellite Imagery" to detect the presence of ships in an image. The dataset is publicly available on Kaggle. The results indicate adoption of transfer learning and data augmentation yields a successful detection of ships with an accuracy of more than 99%. Similarly, exploring different deep learning models for this task provide results with high accuracy for less training time.

Language: en

Keywords

CNN; Marine traffic; Maritime monitoring; Optical satellite imagery; Remote sensing