Multifeature Fusion Neural Network for Oceanic Phenomena Detection in SAR Images

Area of Science:

• Marine remote sensing
• Geophysics
• Artificial intelligence

Background:

• Oceanic phenomena detection in Synthetic Aperture Radar (SAR) images is crucial for fishery, military, and oceanographic applications.
• Traditional methods using handcrafted features and thresholds exhibit poor generalization.
• Existing deep learning methods often focus on detecting only a single type of oceanic phenomenon.

Purpose of the Study:

• To develop an efficient and accurate method for detecting multiple oceanic phenomena and their information in large volumes of SAR images.
• To address the limitations of traditional and single-phenomenon deep learning detection methods.

Main Methods:

• A convolutional neural network (CNN) based approach is proposed for oceanic phenomena detection in SAR images.
• Utilizes ResNet-50 for multilevel feature extraction.
• Incorporates an atrous spatial pyramid pooling (ASPP) module for multiscale feature extraction.
• Fuses multilevel and multiscale features for enhanced detection.

Main Results:

• The proposed method achieves 91% accuracy on a dataset of oceanic phenomena derived from Sentinel-1 satellite SAR images.
• Demonstrates improved generalization ability compared to traditional methods.

Conclusions:

• The developed CNN-based method effectively detects multiple oceanic phenomena in SAR images.
• The feature fusion strategy enhances detection accuracy and efficiency for large-scale SAR data analysis.