Abstract: Apples are the most widely consumed fruits in dominant agricultural products with the largest planting area in China. However, mechanized pruning of apple trees is still lacking, particularly on the automation and intelligence, due to high intensity of manual labor in current China. Artificial intelligence techniques for pruning can be expected to greatly improve the photosynthetic efficiency of fruits, thereby optimizing the transport and distribution of nutrients in fruit trees for better yield and quality. In this study, an intelligent decision support system for pruning apple trees was established using the local point cloud and BP neural network, according to tree shape and artificial intelligent decision making framework. The first fruiting fusiform apple trees were selected as experimental subjects, in order to obtain the actual conditions of tree branches in an apple. The data collection site was selected as the Bakou Fruit Tree Experimental Garden in Beijing, China, and the collection time was autumn 2019. A MESA SR4000 Time of Flight (ToF) depth camera was selected as the data acquisition device. 15 fruit trees with regular shapes were set for data collection, 8 different angles for each tree from which to capture the branch part, and finally, a total of 120 datasets were collected. 5 feature values were used to represent the spatial morphology of apple tree branches, where the local point cloud segmentation technique was selected for the 3D skeleton extraction of branches. Specifically, a Harris corner-point detection was conducted to generate the initial candidate points, then to filter the neighboring candidate points by cohesive hierarchy, and thereby extract the 3D skeletal key points by depth hierarchy analysis further to establish the spatial vectors of branches using the line coverage, as well as obtain the 3D spatial morphological feature data of apple tree branches. Thus, a 3D skeletal map of branches was achieved to realize the digital simulation of the real tree branches. A BP neural network using pruning decision was then established to discriminate the competing branches and disturbing branch types (dorsal, centripetal and competing branches) with 3D skeletal maps, while determining whether the branch was cut or not. The learning rate of the system reached 0.85 after training and testing, when the implicit layer node was 10. The pruning decision scheme has a good degree of discrimination for dorsal and centripetal branches, with an accuracy rate of more than 90%. The overall detection rate of interference branches was 85.71%, indicating a high standard level. Taking Python as the programming language, and Django and TensorFlow as the deep learning framework, a new system was developed with the functions of 3D bone extraction of tree branches, pruning apple tree, and information query of apple tree pruning. The system can realize the digital processing and intelligent pruning of apple trees, providing scientific and reliable tools for mechanized pruning of fruit trees.