DFENet: A Novel Dual-Path Feature Extraction Network for Semantic Segmentation of Remote Sensing Images.

Area of Science:

• Geoscience
• Remote Sensing
• Computer Vision

Background:

• Semantic segmentation of remote sensing images (RSIs) is crucial for geoscience.
• Existing deep learning models struggle with efficient feature fusion and neglect frequency-domain information, leading to semantic confusion and blurred boundaries.
• Addressing these limitations is vital for accurate analysis of complex remote sensing scenes.

Purpose of the Study:

• To propose a novel dual-path feature extraction network (DFENet) for improved semantic segmentation of RSIs.
• To enhance feature fusion capabilities and incorporate frequency-domain information.
• To overcome the limitations of existing architectures in handling complex remote sensing data.

Main Methods:

• Developed a dual-path module (DPM) for extracting global and local features.
• Integrated four feature extraction strategies in the global path for multi-granularity feature extraction.
• Designed a frequency-domain feature extraction block (FFEB) using discrete Wavelet Transform (DWT) to capture high- and low-frequency components.

Main Results:

• DFENet achieved superior segmentation performance compared to state-of-the-art methods.
• The proposed method obtained a mean intersection over union (mIoU) of 83.09% on the ISPRS Vaihingen dataset.
• A mIoU of 86.05% was achieved on the ISPRS Potsdam dataset, demonstrating effectiveness.

Conclusions:

• The proposed DFENet effectively addresses challenges in feature fusion and frequency-domain information utilization for RSIs.
• The integration of DPM and FFEB significantly enhances semantic segmentation accuracy.
• DFENet represents a promising advancement for remote sensing image analysis and geoscience research.