Seismic Risk Classification of Building Clusters Using MST Clustering and UAV Remote Sensing

Area of Science:

• Remote Sensing
• Geospatial Analysis
• Structural Engineering

Background:

• Seismic capacity assessment of houses critically depends on accurate building structure type identification.
• Conventional remote sensing methods often overlook spatial similarity, relying solely on individual house image features, leading to suboptimal classification accuracy.

Purpose of the Study:

• To propose and evaluate a minimum spanning tree (MST) house clustering method that leverages spatial similarity for improved house structure type classification.
• To enhance the accuracy of seismic capacity assessment by refining the classification of simple, brick-concrete, and frame houses.

Main Methods:

• Utilized geometric characteristics of residential buildings to compute the Gestalt factor for visual distance.
• Constructed a Delaunay triangular mesh to generate a house proximity map, with MST weighted by visual distance.
• Employed support vector machine (SVM) for final classification based on geometric, textural, height, and spatial distribution features.

Main Results:

• The MST clustering method achieved high classification accuracy: 95.4% for brick-concrete houses and 93.4% for simple houses.
• Classification accuracy for frame-structure buildings improved to 87% compared to single-family-based methods.
• The Kappa coefficient increased to 0.89, demonstrating a significant overall improvement in classification performance.

Conclusions:

• The proposed MST clustering method, incorporating spatial similarity, substantially improves the accuracy of remote sensing-based house structure type classification.
• This approach offers a promising new direction for enhancing seismic capacity assessment through more precise building classification.
• Spatial similarity is a key factor that can be effectively utilized in classifying building structure types for disaster risk reduction.