Image semantic understanding algorithm based on three-layer LSTM (Long Short Term Memory) promotion network

The invention belongs to the field of image description, and particularly relates to an image semantic understanding method based on a three-layer LSTM (Long Short Term Memory) promotion network. According to an image semantic understanding algorithm, deep semantic information of a picture is mined by generating character description for the picture, the development of the field depends on the development of a computer vision technology and a natural language processing technology at the same time, and a function of generating the character description by the picture is mainly realized by using a coding and decoding framework at present. However, even if the current image semantic understanding algorithm can completely express the content in the image, the expressed content is often not fine enough, and the generated vocabularies cannot accurately express the content, the attribute and the mutual relation of the image. Therefore, an image semantic understanding algorithm based on a coding and decoding framewo