To address the inefficiency and high error rates of manual inspection of three-phase information (batch number, production date, and expiry date) on pharmaceutical packaging in traditional production processes, this study designs an automated pharmaceutical information recognition system based on Optical Character Recognition (OCR). The system aims to enhance quality control and ensure regulatory compliance. It integrates innovative hardware configurations with improved software technologies. On the hardware side, a detection platform was designed to accommodate pharmaceutical packages of various specifications, leveraging photometric stereo technology to capture high-quality embossed character images. On the software side, the text detection and recognition algorithms were improved. The enhanced DBNet algorithm incorporates deformable convolution modules and feature pyramid enhancement modules to improve detection capabilities for complex text scenarios, while the text recognition module integrates deformable attention mechanisms to enhance the distinction of complex characters. Experimental results show that the system achieves an accuracy of 89.0% when processing 1200 low-quality pharmaceutical information images, an improvement of 3.7 percentage points compared to the traditional ABINet model. Additionally, in real production line testing, the system achieved a detection accuracy of 98%-100% for five types of pharmaceutical packaging, with an average detection time approximately one-quarter that of manual inspection. This system significantly improves the efficiency and accuracy of pharmaceutical three-phase information detection and recognition, overcoming the limitations of traditional methods and providing reliable technical support for quality regulation in pharmaceutical production.
Design of an OCR-Based System for Recognition of Pharmaceutical Information[J].
China Medical Devices.
https://doi.org/10.3969/j.issn.1674-1633.20241689