PP-LinkNet: Improving Semantic Segmentation of High Resolution Satellite Imagery with Multi-stage Training

TL;DR

This paper introduces PP-LinkNet, a novel deep neural network with multi-stage training and transfer learning, significantly improving semantic segmentation accuracy for high-resolution satellite imagery in road and building footprint extraction tasks.

Contribution

The paper presents a two-stage transfer learning approach leveraging noisy OSM data and introduces PP-LinkNet with focal loss and context modules for enhanced segmentation performance.

Findings

Achieved 78.19% meanIoU on SpaceNet building dataset.

Attained 67.03% and 77.11% road topology metrics on SpaceNet and DeepGlobe datasets.

Demonstrated improved robustness and accuracy over existing methods.

Abstract

Road network and building footprint extraction is essential for many applications such as updating maps, traffic regulations, city planning, ride-hailing, disaster response \textit{etc}. Mapping road networks is currently both expensive and labor-intensive. Recently, improvements in image segmentation through the application of deep neural networks has shown promising results in extracting road segments from large scale, high resolution satellite imagery. However, significant challenges remain due to lack of enough labeled training data needed to build models for industry grade applications. In this paper, we propose a two-stage transfer learning technique to improve robustness of semantic segmentation for satellite images that leverages noisy pseudo ground truth masks obtained automatically (without human labor) from crowd-sourced OpenStreetMap (OSM) data. We further propose Pyramid Pooling-LinkNet (PP-LinkNet), an improved deep neural network for segmentation that uses focal loss, poly learning rate, and context module. We demonstrate the strengths of our approach through evaluations done on three popular datasets over two tasks, namely, road extraction and building foot-print detection. Specifically, we obtain 78.19\% meanIoU on SpaceNet building footprint dataset, 67.03\% and 77.11\% on the road topology metric on SpaceNet and DeepGlobe road extraction dataset, respectively.