Abstract: In order to manage urban trees in an efficient way, this paper studies the classification of urban street trees. Aiming at the complex problems such as learning multi-classification model optimization of urban street trees, an adaptive deep learning method is proposed by considering the multi-classification characteristics of urban roadside trees. The feature engineering method based on random forest learning is adopted to calculate and analyze the feature importance of urban roadside trees, and the unimportant features are discarded by recursive feature elimination method. Moreover, to improve the performance of multi-classification learning algorithm for urban street trees, an adaptive deep learning method is further constructed on the basis of urban tree feature learning. Furthermore, the proposed deep learning model is evaluated and improved by cross-validation and parameter search methods. The simulation results show that our proposed algorithm has superior performance and effectively to solve the problem of accuracy and generalization of multi-classification of urban street trees.

Key words: urban street trees classification, data representation, feature engineering, deep learning model, model evaluation and optimization