China’s rapid and extraordinary urbanisation has resulted in a dichotomy between the urban form of historic districts and that of new urban areas. Unlike the new planning of new urban areas, the optimisation of the thermal environment in historic districts is achieved through an important means of regulating the shading effectiveness of trees in street canyons, without destruction of current buildings in existing districts. In this paper, the sky view factor, an index for evaluating the shading capacity of street canyons, is used to calculate the shading efficiency of trees and reveal their influence mechanism by using a convolutional neural network (SegNet) to segment the Baidu street view images. The results show that: (1) tree shading in Harbin’s historic districts has a significant impact on the thermal environment of street canyons, with an average shade effectiveness of 56.3%; (2) based on geospatial autocorrelation analysis and a typological reconstruction of street canyons, the study reveals that tree shading has significant spatial aggregation characteristics in historic districts and proposes guidelines for the planning and design of different types of street canyon trees. The study provides important data and strategic support for optimising the thermal environment of urban historic districts in Northeast China.