A natural scene character recognition method for warehouse shelf signboard character recognition

Abstract

The invention provides a natural scene character recognition method for warehouse shelf signboard character recognition. The natural scene character recognition method at least comprises the followingsteps of building a to-be-recognized signboard text detection network, wherein the specific structure of the to-be-identified signboard text detection network is 13 convolutional layers from VGG-16,a full convolutional network with convolutional layers being additional convolutional layers for extracting 10 features, and six text box layers connected to six intermediate convolutional layers; andreserving 13 convolutional layers of the VGG-16, and fully linking and replacing the two full-connection layers formed by the VGG-16 with the two convolutional layers adopting a parameter downsampling principle. According to the natural scene character recognition method for the warehouse shelf signboard character recognition provided by the invention, the natural scene character recognition accuracy, the precision rate and the recall rate in a logistics warehouse environment can be relatively high, and meanwhile, the natural scene character recognition method also has very good efficiency.

Description

technical field [0001] The present invention relates to the technical field of character recognition, in particular, to a natural scene character recognition method for character recognition of warehouse shelf signboards. Background technique [0002] Natural scene text recognition technology is different from traditional OCR (Optical Character Recognition) technology and can be divided into two parts: text detection and text recognition. Text detection has the following methods: In the CTPN scheme, the BLSTM module is used to extract the image context features where the characters are located to improve the recognition accuracy of text blocks. In RRPN and other schemes, the text box is marked in the form of BBOX + direction angle value, and the rotatable text area candidate box is generated in the model, and the inclination angle of the text line to be tested is found during the frame regression calculation process. In programs such as DMPNet, a quadrilateral (non-rectangu...

AI Technical Summary

Problems solved by technology

[0008] With the development of the logistics industry, smart logistics will become the main direction of the industry's development. Due to the gradual expansion of the logistics scale, the number of shelf signs in the logistics warehouse is increasing, and the existing natural scene text recognition technology is lacking in the application of logistics. In the warehouse environment, and the accuracy rate, precision rate and recall rate are not very ideal

Images