The Effectiveness of Identifying Residential Housing using Image Recognition by Artificial Intelligence

Abstract

Residential housing identification based on satellite imagery has become an important approach in supporting urban planning, disaster management, and regional mapping. This study evaluates the effectiveness of settlement recognition techniques using high-resolution imagery and artificial intelligence (AI) models, specifically deep learning methods based on convolutional neural networks (CNN) and object segmentation. The main factors that affect identification accuracy include image spatial resolution, preprocessing quality, training data diversity, and the geographic complexity of the observed area. Based on the analysis results, the use of high-resolution imagery combined with image recognition by AI such as Google Gemini and ChatGPT can produce an accuracy of 68.4% in calculating the number of buildings. This value tends to be low to achieve a high level of accuracy in calculating the number of buildings therefore it is not recommended to calculate the number of buildings accurately, but it can be used to determine housing availability in a range of values. However, to analyze building density, AI can successfully generate complex images of building density according to the conditions of the given image. AI can be used to aid urban planning while emphasizing the importance of selecting data sources, careful preprocessing techniques, and adaptive machine learning models to improve the effectiveness of settlement recognition, especially in areas with complex spatial structures in the fields of regional and urban planning and disaster management.