Recognizing the centrality of ship detection and segmentation based on optical remote sensing images in the military and civilian domains, the sea surface remote sensing field is receiving a lot of attention. The deep learning-based ship detection approach can also extract vessel position and category information, crucial for maritime surveillance. Unfortunately, less research has been done in this area. The majority of known approaches for spotting ships perform object detection but don't perform semantic segmentation. Furthermore, earlier studies included limitations such as the inability to detect small ships or ships of diverse sizes and the higher rate of false positives. Convolutional Neural Networks (CNNs) benefits are also significantly reduced due to the complexities of SAR images. Then there are a few models that can't discern the difference between items that seem like ships, and these false alarms lower the total accuracy rate. The available models have also faced trouble while identifying and segmenting ships with a complicated background and dealing with light variations. Furthermore, some existing models performed somewhat worse than other state-of-the-art frameworks. Some models have a higher cost of computation. This paper uses RetinaNet to provide the ship detecting algorithm. Instance Segmentation was also included in the model. The proposed model was created by integrating these two strategies. RetinaNet is a fantastic one-stage object detection model that can precisely work with dense and small-scale objects. The Feature Pyramid Network (FPN) and Focal Loss are two improvements over conventional one-step object detection models. Then basically, an instance segmentation algorithm has been implemented in place of the object detector to increase RetinaNet's performance by removing background better. The proposed model can work more precisely than existing models due to these adjustments. We achieved a promising result using the described techniques on the Airbus dataset from Kaggle, with Average Precision (AP) score of 65.7 and APs of 51.8, respectively.
Ship Detection from Satellite Imagery Using RetinaNet with Instance Segmentation
Lect. Notes in Networks, Syst.
Proceedings of International Conference on Recent Trends in Computing ; Chapter : 54 ; 627-639
2023-03-21
13 pages
Article/Chapter (Book)
Electronic Resource
English
Optical remote sensing images , Semantic segmentation , Convolutional neural networks , Synthetic aperture radar , RetinaNet , Feature pyramid network , Focal loss , Instance segmentation , Average precision Engineering , Computational Intelligence , Artificial Intelligence , Cyber-physical systems, IoT , Professional Computing , Big Data
IS-Net: Improved Ship Detector Based on RetinaNet
IEEE | 2022
|Ship Detection Using High Resolution Satellite Imagery and Space-Based AIS
British Library Conference Proceedings | 2013
|ShipSegNet: A Contour-Driven Instance Segmentation Method for Maritime Ship Segmentation
Springer Verlag | 2025
|