Identifying Counterfeit Medicine in Bangladesh Using Deep Learning

Bilkis Jamal Ferdosi, Mashroor Ahmed Sakib, Md. Sirazul Islam, Joy Dhar

Abstract: The abundance of counterfeit medicine especially in countries like Bangladesh is causing a serious threat to public health. Several invasive approaches and lab-based methods are available but not affordable by the common people neither in terms of money nor in terms of specialist knowledge. In this paper, we proposed a non-invasive alert system to separate approved brands of pharmaceutical companies from fake or banned brands using logo images of the company taken using a mobile phone. Later, if the logo is identified as from an approved brand, then we check if the logo is real or fake. If the logo is not listed in the approved brands or in the list of banned brands or the logo is fake, the system will alert the user about probable fake medicine. We used the pre-trained VGG-16 model through transfer learning and obtained 96% test accuracy in brand recognition and 84% test accuracy in fake detection.