A method of analyzing an image based on a random tree convolutional neural network (CNN) includes: a step in which a computer device inputs a source image into a root node of a tree of which one node corresponds to a CNN; a step in which the computer device delivers the source image to a child node of one node of the tree by classifying the source image in the CNN of the node of the tree; and a step in which the computer device classifies the source image by using a CNN of a terminal node receiving the source image lastly. The tree is randomly branched from a CNN included in the root node while learning a sample image beforehand, and a CNN of the child node of the tree learns a subset of a sample image set learned by a CNN of a parent node of the child node. Furthermore, a forest CNN including a plurality of tree CNNs is able to be used to conduct image analysis according to the principle of majority rule.
展开▼