Wave-Net: A Marine Raft Aquaculture Area Extraction Framework Based on Feature Aggregation and Feature Dispersion for...

Area of Science:

• Marine aquaculture sustainability
• Remote sensing for environmental monitoring
• Deep learning applications in geospatial analysis

Background:

• Marine raft aquaculture monitoring is crucial for sustainability.
• Synthetic Aperture Radar (SAR) offers advantages over traditional methods for aquaculture monitoring.
• Existing deep learning models struggle with SAR image noise and multi-scale structures, leading to inaccurate segmentation of aquaculture areas.

Purpose of the Study:

• To develop an improved deep learning model for accurate semantic segmentation of marine raft aquaculture areas from SAR imagery.
• To address the challenges of speckle noise and multi-scale object detection in SAR images.
• To enhance the accuracy and reliability of marine aquaculture monitoring systems.

Main Methods:

• Proposed a novel Wave-Net architecture comprising feature aggregation and dispersion parts.
• Utilized asymmetric V-shaped subnetworks for extracting multi-scale global and local features.
• Employed asymmetric Ʌ-shaped subnetworks for refining object boundaries.
• Incorporated residual connections and reconstruction losses for feature fusion and parameter optimization.

Main Results:

• The proposed Wave-Net model effectively handled speckle noise and multi-scale structures in SAR images.
• Achieved improved segmentation accuracy by accurately delineating raft aquaculture area boundaries.
• Demonstrated superior performance compared to existing methods, especially with limited training samples.

Conclusions:

• Wave-Net provides a robust solution for semantic segmentation of marine raft aquaculture areas in SAR images.
• The model's ability to refine boundaries significantly improves monitoring accuracy.
• This advancement contributes to more sustainable practices in marine aquaculture.