AI-driven tree detection for enhancing urban livability in sustainable cities

Urban livability is a multifaceted concept encompassing accessibility, well-being, displacement ergonomics, and urban attractiveness. Among its key components green infrastructure plays a crucial role. This research proposes a methodology integrating remote sensing technologies and convolutional neural networks (CNNs) to assess and enhance the understanding of urban vegetation coverage, presenting a two-step approach: green percentage calculation and tree recognition. Aerial orthophotos from public geoportals serve as primary inputs for evaluating the urban green cover, including pixel classification and false-color image extraction. CNNs are employed for automatic tree detection and classification from street-level imagery. The city of Turin was chosen as a case study due to the availability of rich geospatial datasets and urban greening strategies. The estimated green percentages for various districts demonstrated a high degree of agreement with the manually annotated reference data, with an error margin of ±5%. Results demonstrate that CNNs can reliably (with an overall accuracy of 80.8%) classify tree species, despite limitations such as image quality and lack of multispectral data. The discussion highlights automation in environmental monitoring benefits and the potential for improving policy-making and urban planning. Future development will involve multispectral sensors integration and real-time applications.