High-Resolution Aerial Imagery Semantic Labeling with Dense Pyramid Network

Area of Science:

• Computer Vision
• Remote Sensing
• Artificial Intelligence

Background:

• High-resolution aerial image semantic segmentation faces challenges due to intra-class variance and inter-class similarity.
• Deep convolutional neural networks (DCNNs) offer a promising approach by leveraging high-level contextual features.

Purpose of the Study:

• To propose a novel Dense Pyramid Network (DPN) for enhanced semantic segmentation of high-resolution aerial imagery.
• To address challenges in classification ambiguity and class imbalance inherent in aerial image data.

Main Methods:

• The DPN utilizes group convolutions and channel shuffle for effective multi-sensor data processing and feature representation.
• Densely connected convolutional blocks and a pyramid pooling module fuse multi-resolution and multi-sensor features.
• A median frequency balanced focal loss function is introduced to mitigate class imbalance during training.

Main Results:

• The proposed DPN framework demonstrates superior performance on the ISPRS Vaihingen and Potsdam 2D semantic labeling dataset.
• The network effectively extracts and integrates features, improving segmentation accuracy compared to state-of-the-art baselines.

Conclusions:

• The novel Dense Pyramid Network provides an effective solution for semantic segmentation of high-resolution aerial images.
• The proposed methods successfully tackle classification ambiguities and class imbalance, advancing the field of remote sensing image analysis.