基于OCR的药品信息识别系统设计OA

Objective To solve the problems of low efficiency and high error rate of manual detection of the third phase information(batch number,production date and expiration date)of drug packaging in the traditional drug production process,this study designs an optical character recognition system by combining optical image optimization and deep learning algorithm improvement.Methods The hardware detection platform of multi-angle LED light source and industrial camera was built to enhance the image quality of imprinted characters by combining with photometric stereoscopic technology.The software algorithm adopted the improved DBNet detection model(integrated deformable convolution module and feature pyramid enhancement module-feature fusion module architecture)and ABINet recognition model with integrated deformable attention mechanism to improve the text detection and low-quality character recognition capability in complex scenes.Results The recognition accuracy of the system was 89.0%in the test of 1200 low-quality drug images,which was 3.7%higher than the traditional ABINet model.At the same time,in the production line measurement,the detection accuracy of 5 types of drug packaging reached more than 98.5%,the average system detection time was about 1/4 of the average manual detection time.Conclusion Through hardware imaging optimization and algorithm architecture innovation,this study effectively solves the problem of drug information recognition in pharmaceutical companies that is easy to make mistakes and difficult to control.