# Automatized marine vessel monitoring from sentinel-1 data using convolution neural network

**Authors:** Surya Prakash Tiwari, Sudhir Kumar Chaturvedi, Subhrangshu Adhikary, Saikat Banerjee, Sourav Basu  
**arXiv:** [2304.11717](https://arxiv.org/abs/2304.11717)  
**Date:** 2023-04-23  
**Summary page:** https://www.summarizepaper.com/en/paper/2304.11717v1/

## Key points

- Multi-channel synthetic aperture radar (SAR) systems have enhanced surveillance in coastal ocean monitoring and object recognition
- Traditional vessel monitoring methods using the CFAR algorithm lack decision-making capabilities during heavy naval traffic
- A wavelet transformation-based CNN approach is proposed for object recognition from SAR images
- Copernicus Sentinel-1 SAR data was used for sea observation, capturing data throughout the day and night under varying landscape conditions
- Wavelet transformation was applied to denoise the images before passing them to the CNN model for detection
- The study focused on the coastline of India, specifically the Mumbai Colaba port to Navi Mumbai port area on the Arabian Sea
- The wavelet transformation-based CNN model achieved a detection accuracy of 95.46% during heavy naval traffic
- The automated monitoring system has potential for surveillance and recognition of naval objects, including identifying foreign maritime intruders
- Utilizing Copernicus Sentinel-1 SAR data and a wavelet transformation-based CNN approach enhances marine vessel monitoring capabilities

## Summary

The advancement of multi-channel synthetic aperture radar (SAR) systems has greatly enhanced surveillance activities in coastal ocean monitoring and object recognition. Traditional vessel monitoring methods using the Constant False Alarm Rate (CFAR) algorithm lack decision-making capabilities and may not be effective during heavy naval traffic. To address this limitation, a wavelet transformation-based Convolution Neural Network (CNN) approach is proposed for object recognition from SAR images. The study utilizes Copernicus Sentinel-1 SAR data for sea observation, which is capable of capturing data throughout the day and night under varying landscape conditions. The data acquisition was conducted over two unique locations in Indian water bodies – coasts and ports – to experiment with the proposed technique. Prior to passing the SAR data to the CNN model for detection, a wavelet transformation was applied to denoise the images. The study focuses on the coastline of India, specifically the Mumbai Colaba port to Navi Mumbai port area situated on the Arabian Sea. By utilizing this wavelet transformation-based CNN model, a detection accuracy of 95.46% was achieved in recognizing objects from SAR images during heavy naval traffic. This automated monitoring system has significant potential for surveillance and recognition of naval objects, including identifying foreign maritime intruders. Overall, this research demonstrates the effectiveness of utilizing Copernicus Sentinel-1 SAR data and a wavelet transformation-based CNN approach for automatized marine vessel monitoring. The combination of advanced radar technology and deep learning algorithms offers a promising solution for enhancing surveillance capabilities in coastal areas and improving maritime security.