The reasonable structure of fruit tree canopy is beneficial to the effective distribution of illumination, which has vital significance to enhance the fruit yield and quality. At present, it is difficult to obtain illumination intensity data in the canopy of fruit trees, and the prediction accuracy is low. In order to study inner canopy illumination distribution, a random forest prediction model was proposed based on canopy profile shadow feature and point cloud color feature. The detailed research methods were shown as follows. Firstly, the spindle “Shanfu 6” apple tree was chosen as the research object and Kinect 2.0 was used to acquire double face point cloud data of tree, and then the complete data was obtained with preprocess. Secondly, the improved space colonization algorithm with growth angle constraint and phyllotaxis adding rules were used to rebuild apple tree 3D model. Finally, the “slice method” was used to cut canopy model every 0.1m in the vertical direction, and then the POV-Ray renderer was used to render shadows layer after layer, meanwhile, light meter was used to obtain illumination intensity data every 0.1m from top to bottom consistently, and the random forest network that with input data of color feature of every layer and output data of relative illumination intensity was built as the apple tree canopy illumination distribution prediction model. The experiment results showed that the proposed method can predict the illumination distribution accurately. The determination coefficient R2 between true value and predicted value was 0.864, and MAPE was 0.236. Random forest regression model can be used as an efficient method for prediction of canopy illumination distribution, and it can provide reference for fruit tree pruning and plastic research.