Title:
NEURAL NETWORK-BASED DOCUMENT CLASSIFICATION METHOD AND APPARATUS, AND DEVICE AND STORAGE MEDIUM
Document Type and Number:
WIPO Patent Application WO/2021/000411
Kind Code:
A1
Abstract:
A neural network-based document classification method and apparatus, and a device and a storage medium, relating to the technical field of artificial intelligence-based image processing. The method comprises: receiving a first page image and a second page image (201); invoking a first convolutional neural network and a second convolutional neural network to extract text features and image features respectively (202); combining the text features and the image features to generate a hybrid document feature (203); invoking a multilayer perceptron and inputting the hybrid document feature to obtain an output predicted value (204); and on the basis of a predicted value, determining whether the first page image and the second page image belong to the same document (205). According to the method, by using two convolutional neural networks and a multilayer perceptron in combination, the two aspects, i.e., text features and image features, in scanned text images, are combined, so that a large batch of document images can be automatically classified to make the classification process more reasonable and efficient, thereby improving classification efficiency, and two performances, i.e., accuracy and consistency, can be significantly improved.
Inventors:
WANG JIANZONG (CN)
HUI YANFEI (CN)
HAN MAOKUN (CN)
HUI YANFEI (CN)
HAN MAOKUN (CN)
Application Number:
PCT/CN2019/103450
Publication Date:
January 07, 2021
Filing Date:
August 29, 2019
Export Citation:
Assignee:
PING AN TECH SHENZHEN CO LTD (CN)
International Classes:
G06F16/93
Foreign References:
| CN108984706A | 2018-12-11 | |||
| CN109582794A | 2019-04-05 | |||
| CN109344815A | 2019-02-15 | |||
| CN108763325A | 2018-11-06 | |||
| US20150178563A1 | 2015-06-25 |
Attorney, Agent or Firm:
SL INTELLECTUAL PROPERTY CO., LTD. (CN)
Download PDF: