基于深度学习的药盒药名识别技术研究

Research on Recognition Technology of Medicine Box Drug Name Based on Deep Learning

在智能化药房中,为完成对药品存/取操作,需要为机器人配备视觉系统来实现对目标药品的定位与名称识别。对基于深度学习的文本检测与文字识别方法进行理论分析与实验,采用CRAFT算法检测药品文本区域,根据药名文本的特征检出并截取药名区域;基于LSTM的Tesseract-OCR对药名进行文字识别。对202个实物药盒进行的识别实验表明,对药名区域检测截取的准确率为98.02%;对药品名称的识别中,常规字体的准确率可以达到91.09%,对进一步研究提高识别效果具有参考意义。

In the intelligent pharmacy,in order to complete the storage/retrieval of drugs,it is necessary to equip the robot with vision system to realize the target drug location and name recognition.The method of text detection and character recognition based on deep learning is analyzed and tested.The drug text area is detected by CRAFT algorithm,and the drug name area is intercepted by the feature of the drug name text Character recognition of drug names by Tesseract-OCR based on LSTM.The experiments on 202 real medicine boxes show that the accuracy of intercepting the medicine name region is 98.02%,and the accuracy of the conventional font can reach 91.09%in the medicine name recognition,it has reference significance for further research and improvement of recognition effect.