Model, model training method and device, character sequence recognition and device

Abstract

The application relates to the technical field of artificial intelligence, and relates to image processing or character sequence processing technology under an artificial intelligence network, in particular to an identification model capable of identifying a character sequence, which comprises an encoding module, which can be used to obtain context features according to input data; a first decoding module, which can be used to obtain an identified character sequence according to the context features, wherein the first decoding module comprises: a character position prediction module, which can be used to obtain a feature map according to the context features, the feature map reflecting position information of characters in the character sequence; and a first sequence processing module, which can be used to obtain the character sequence according to the context features and the feature map. The identification model can be trained in a manner of knowledge distillation by using a sequence decoder based on serial attention, so that the identification model has relatively high running efficiency and also has high character recognition precision.

Images