Lightweight multiscale information aggregation network for land cover land use semantic segmentation from remote...

Area of Science:

• Remote Sensing
• Computer Vision
• Geospatial Analysis

Background:

• Land Cover and Land Use (LCLU) segmentation is crucial for environmental monitoring, urban planning, and disaster management.
• Traditional models struggle with real-time processing and deployment on resource-constrained devices due to high computational demands.

Purpose of the Study:

• To develop a lightweight neural network for efficient and accurate LCLU segmentation.
• To enable real-time LCLU analysis on resource-constrained devices.

Main Methods:

• Integration of dense dilated convolutions and pyramid depthwise convolutions for multiscale feature extraction.
• An encoder-decoder architecture with dense connections to aggregate spatial and contextual information.
• Evaluation on NITRDrone and UDD6 datasets.

Main Results:

• Achieved 94.8% segmentation accuracy.
• Significantly reduced parameter count compared to state-of-the-art methods.
• Demonstrated feasibility for real-time LCLU analysis on low-power devices.

Conclusions:

• Lightweight neural networks offer scalable and efficient solutions for remote sensing image processing.
• The proposed model advances practical applications in geospatial analysis.