A Preliminary Study on Remote Sensing-Based Detection of Vacant Houses in Urban Areas
Shota Tsujino (1)*, Kiichiro Kumagai (1),Kazuki Hatao (2)

(1) Setsunan University, Japan
(2) HawksMap Co., Ltd., Japan

Abstract

In Japan, the population is declining at a rapid pace due to a sharp drop in the birth rate and the progression of an aging society. As a result, the importance of continuously monitoring urban structures, living environments, and related social dynamics over the long term has increased significantly. Among various influencing factors, accurately identifying the spatial distribution of vacant houses is expected to be an effective and practical means of observing underutilized, low-utilized, or abandoned spaces in diverse urban environments. However, in field surveys aimed at accurately identifying vacant houses, various issues such as physical access restrictions, obstructed views due to adjacent buildings or vegetation, safety concerns, and legal privacy restrictions often make it difficult or impossible to visually confirm the exterior of buildings. To comprehensively address these challenges, we have focused on aerial images captured by unmanned aerial vehicles (UAVs, commonly known as drones) and proposed a method to effectively identify the condition of the surrounding environment of residential properties as an objective and quantitative indicator of vacant or underutilized residential properties by leveraging deep learning techniques for image recognition. Considering the potential application of satellite imagery for large-scale and efficient monitoring, this study first performed interpolation processing on UAV images as a preliminary step and compared the differences in identification results due to variations in spatial resolution. Furthermore, image-based classification was performed on all buildings within the designated field survey area, and the classification results of the original high-resolution UAV images and the interpolated low-resolution images were compared in detail. Through this detailed comparison, this study verified the impact of image resolution on classification accuracy and confirmed the potential of this approach as an expandable and cost-effective method for supporting large-scale identification and continuous monitoring of vacant house distribution.

Keywords: Vacant houses,UAV imagery,Deep learning,Spatial resolution,Image recognition

Topic: Topic B: Applications of Remote Sensing