Urban tree diversity is responsible for many ecosystem services. Quantifying the level and spatial pattern of tree diversity is crucial for the assessment of ecosystem services. However, previous quantitative studies on urban tree diversity are mainly based on field surveys, which are highly expensive and restricted by poor efficiency. In this study, we firstly tested the applicability of Spectral Variation Hypothesis (SVH) and Productivity Hypothesis in urban areas. Then, we developed a new approach for quickly and efficiently quantifying tree diversity using remote sensing imagery. This new approach described the relationship between urban tree diversity and spectral heterogeneity based on Neural Network Model, which could be used to calculate the Shannon-Wiener index and the Simpson index of urban trees. This study provides a new direction for studying urban tree diversity, and our results are critical basic data for studying ecosystem services of urban tree diversity.