A Review on Recent Deep Learning-Based Semantic Segmentation for Urban Greenness Measurement

Area of Science:

• Environmental science
• Computer science
• Urban planning

Background:

• Accurate measurement of urban green space (UGS) is essential for effective landscape analysis and urban planning.
• Traditional methods for UGS measurement face limitations in accuracy and efficiency.
• Recent advancements in deep learning (DL) offer potential solutions for enhanced UGS analysis.

Purpose of the Study:

• To review recent technological breakthroughs in deep learning (DL)-based semantic segmentation for urban green space measurement.
• To explore quantitative greenness measures and their integration with semantic segmentation techniques.
• To illuminate the evolution of UGS measurement and segmentation tasks for advanced landscape analysis.

Main Methods:

• Review of deep learning (DL) models for semantic segmentation applied to UGS.
• Categorization of quantitative greenness measures into plan view- and perspective view-based methods (e.g., Land Class Classification (LCC), Green View Index (GVI)).
• Exploration of performance metrics and public datasets for UGS measure construction.
• Discussion of unsupervised domain adaptation (UDA) for aerial imagery.

Main Results:

• Accurate semantic segmentation is crucial for fine-grained greenness measures and qualitative landscape evaluation.
• Deep learning models significantly improve the efficiency and accuracy of UGS analysis.
• Unsupervised domain adaptation (UDA) helps address challenges like scale changes and limited labeled data in aerial images.

Conclusions:

• Deep learning-based semantic segmentation is indispensable for accurate urban green space measurement and landscape analysis.
• Further research into DL model explainability and UDA techniques is needed for robust UGS quantification.
• This review provides insights into DL advancements for researchers in urban green space studies.