Urban Land Use Classification in Metro Manila using Deep Learning

Abstract. Land use maps play a pivotal role in proper land use planning, as they represent not just the physical characteristics of the land but also the various socio-economic activities that it undertakes. Proper land use planning and monitoring ensure that urbanization and development is sustainable. In the Philippines, the Department of Human Settlements and Urban Development (DHSUD) has land use guidelines as part of the Comprehensive Land Use Plan (CLUP) preparation in each city or municipality. While remote sensing technologies have made mapping easier, it can only see building roofs, which makes land use monitoring a tall order. Land use monitoring often requires conventional efforts, including ground validation surveys to account for the rapid changes in land use patterns. This study developed a methodology to rapidly classify urban land use in Metro Manila. The training dataset is composed of images that were scraped from Google Street View from selected points of interest in the region. A neural network named EfficientNetV2 was used for training due to its great accuracy in image classification tasks while being fast and lightweight. The trained model achieved a 60.4% overall accuracy for the testing dataset. The class f-1 score ranges from 0.40 for the Government Office class up to 0.83 for the Parks class. The developed methodology exhibited its capability in rapidly creating land use maps, especially in urban areas, highlighting its potential to be used in urban planning applications and research.