The invention discloses a substation property segmentation method based on a convolutional neural network. A substation image database which contains seven types of images is established, the images are subjected to attribute table construction and manual labeling semantic segmentation, attribute classification and attribute segmentation convolutional neural networks are pre-trained and learnt, and the images are subjected to attribute classification and attribute segmentation through the convolutional neural network. The substation property segmentation method based on the convolutional neural network can effectively prevent and eliminate power transmission faults and guarantees safe and smooth power supply; and meanwhile, the pressure of labor power monitoring can be effectively alleviated to realize intelligent monitoring in a real sense. When the substation image database is established, a great quantity of tests is carried out, and results indicate that the proposed novel technology based on the depth convolutional neural network is high in practicability and feasibility.