Machine-made invoice character recognition method based on pulse active learning

Abstract

The invention discloses a machine-made invoice character recognition method based on pulse active learning, and the method comprises the steps: constructing deep pulse neural network models ResNet-18 and CIFARNet which can be directly trained, and designing the specific steps of pulse active learning. The active learning is used for selecting effective samples capable of providing more information for the model, so that the model is trained with the minimum data volume and the best effect is achieved. Character recognition application is carried out on a pulse active learning algorithm in a project of recognizing machine-made invoices, character parts of the machine-made invoices are extracted, single character segmentation is carried out on extracted Chinese, English and numbers, finally segmented single character samples are input into a model for training, and the model can screen samples with the maximum information amount. Only samples selected by the model are manually marked.

Description

technical field [0001] The invention belongs to the technical field of artificial intelligence and relates to a character recognition method for machine-printed invoices based on pulse active learning. Background technique [0002] Machine-printed bill images are ubiquitous in daily life. For users, there is a large amount of bill information collection and processing work every day. Traditional manual input of information is inefficient and expensive; training neural networks for bill recognition requires manual labeling of each character cut out, costly Expensive and time-consuming. [0003] In machine learning tasks, due to the high cost of data labeling, researchers are faced with the problem of how to obtain the most effective learning model with the least amount of samples. In response to this problem, the academic community has proposed the research direction of active learning. Active learning formulates a "selection strategy" for specific tasks, selects the sample...

AI Technical Summary

Problems solved by technology

[0010] The purpose of the present invention is: in the character recognition task of machine-printed invoices, it is extremely expensive to manually mark the image data of each individual character that is cut. In order to solve the problem of high cost of training the pulse neural network model, we will actively learn the The idea is to filter representative training samples in the spiking neural network, so as to reduce the cost of training the spiking neural network model and use as little data as possible to make the model achieve better performance

Images