Related Experiment Video
Updated: Aug 11, 2025
04:48
Application of Deep Learning-Based Medical Image Segmentation via Orbital Computed Tomography
Published on: November 30, 2022
2.9K
Ocean oil spill detection from SAR images based on multi-channel deep learning semantic segmentation
Rogelio Hasimoto-Beltran1, Mario Canul-Ku1, Guillermo M Díaz Méndez2
1Centro de Investigación en Matemáticas (CIMAT), Jalisco S/N, Col. Valenciana, Guanajuato 36023, Guanajuato, Mexico.
Marine Pollution Bulletin
|February 3, 2023
Summary
This study introduces a novel Multi-channel Deep Neural Network (M-DNN) for accurate oil spill detection using Synthetic Aperture Radar (SAR) images. The M-DNN model achieves 98.56% accuracy, significantly improving marine pollution monitoring.
Area of Science:
- Marine pollution monitoring
- Remote sensing for environmental applications
- Deep learning for image segmentation
Background:
- Marine ecosystems face significant threats from oil spills, particularly from the offshore oil and gas industry.
- Accurate and timely detection of oil spills is crucial for mitigating environmental damage.
- Existing deep neural network (DNN) models for oil spill detection have limitations in handling complex oil shapes and varying environmental conditions.
Purpose of the Study:
- To develop a new Multi-channel Deep Neural Network (M-DNN) segmentation model for improved oil spill detection and monitoring.
- To create a novel and effective Synthetic Aperture Radar (SAR) image dataset for training and validating the M-DNN model.
- To enhance the reliability and timeliness of oil spill forewarnings.
Main Methods:
- Development of a new Multi-channel SAR Image Database using ENVISAT-ASAR images from various oil spill incidents.
- Implementation of a Multi-Channel DNN model integrating U-net and ResNet architectures.
- Training and transfer learning of the M-DNN model using a 3-channel input image (radiometric values, wind speed estimation, and variance).
Main Results:
- The M-DNN model achieved a pixel-level classification accuracy of 98.56%, the highest reported for DNN models in oil spill detection.
- Utilizing 2 or 3 input channels in the M-DNN significantly improved classification accuracy.
- The M-DNN model demonstrated a fast training convergence rate, approximately 14 times faster than previous methods.
Conclusions:
- The proposed M-DNN model offers a highly accurate and efficient solution for detecting and monitoring oil spills using SAR imagery.
- The developed multi-channel SAR dataset and M-DNN architecture effectively address challenges posed by complex oil spill shapes and environmental factors.
- This work represents the first multi-channel DNN-based scheme for classifying oil spills at various scales, advancing marine pollution detection capabilities.
