Image Chinese description method based on local features and second-order recurrent neural network

The invention discloses an image Chinese description method based on local features and a second-order recurrent neural network. The image Chinese description method can be applied to image understanding scenes such as automatic commodity description and infant early education. The method is characterized by comprising the following steps: constructing a target detection module to obtain image local features, so that a generated description contains an object in an image as much as possible; a local feature enhancement module is constructed to extract and refine local features, and interference is reduced; and a second-order recurrent neural network module and an attention module are constructed to decode the local features to generate Chinese description of the image, and the second-order recurrent neural network can establish a better long sequence dependency relationship, so that the generated description sentence is more natural and smoother. Experimental results show that the method has the characteristi