Abstract:
In Bangladesh, counterfeit medications pose a serious threat to public health since they reduce treatment ability and erode public confidence in the medical establishment. This thesis focuses on creating a deep learning-based system for detecting counterfeit medications with the goal of improving the precision and dependability of detecting counterfeit pharmaceutical items. Using a deep learning framework, the study processes and analyzes these photographs with the goal of identifying authentic products from counterfeit ones by comparing minute variations in logos of pharmaceutical companies in Bangladesh. These visual cues are essential for spotting fakes since they frequently have dangerous or inaccurate components. This study's foundation is a speciallyassembled dataset of 1,543 pictures, divided between 879 genuine and 664 fake logos. The photos serve as a varied base for training and testing the detection system. They are taken from official pharmaceutical channels and enforcement seizures. Carefully dividing the dataset into separate sets for testing, validation, and training allows for a thorough assessment of the system's performance on untested data. The detection system's performance is assessed using measures like mean Average Precision (mAP), F1-score, accuracy, precision, and recall. These measures measure the system's efficacy in effectively identifying counterfeit products while reducing false positives, which gives regulatory and health organizations dependable support. The study's findings demonstrate the potential benefits of using cutting-edge data analysis methods to important public health applications. The suggested approach can play a major role in protecting public health by effectively recognizing fake medications. In addition, this study establishes the foundation for future developments in the use of technology to address health-related issues, and it may also broaden its approach to other industries where the use of fake goods is problematic
