基于DBNet和CRNN算法的端到端企业实体识别

End-to-end enterprise entity identification based on the DBNet and CRNN algorithms

随着深度学习技术的发展,文字识别与自然语言处理近年来受到广泛关注.结合文字识别与自然语言处理技术解决传统方法无法处理的问题,成为企业提高自身竞争力的重要利器.自然场景文字识别分为文字的检测和识别,两者缺一不可.本研究针对传统算法存在准确率低、识别速度慢及模型不轻量化等问题,提出一种基于DBNet的检测算法,结合CRNN的识别算法,辅以CTC loss来实现端到端的企业实体识别.此外,增加命名实体识别模块,提升了识别的准确度.在实验阶段,选择准确率(Precision,P)和识别速率(False Alarm,FA)作为评价指标,实验结果表明,本算法在数据集上,有较高的准确率和较快的识别速率,验证了所提出的改进方法并具有较好的效果.

With the development of deep learning technology,word recognition and natural language processing have received wide attention in recent years.The technology combining text recognition and natural language processing technology to solve the problem of traditional methods has been an important tool for enterprises to improve their own competitiveness.Text recognition in natural scenes is divided into text detection and recognition,and both algorithms are indispensable.For traditional algorithms,there’re problems such as low accuracy,slow recognition speed and not lightweight model.For these problems,in this paper,we proposed a DBNet-based detection algorithm,combined with a CRNN recognition algorithm,supplemented by CTC loss to achieve end-to-end enterprise entity recognition.In addition,we added the named entity recognition module to improve the accuracy of recognition.In the experimental stage,the accuracy(Precision,P)and identification rate(False Alarm,FA)were selected as the evaluation indicators.According to the experimental results,this algorithm has high accuracy and fast identification rate on the dataset,which verifies that the proposed improvement method has good results.