Pushing the Boundaries of Pharmaceutical Drug Classification with Advanced Deep Learning Techniques

Capsules and tablets are the best and most feasible method of storing medicines. Meanwhile people are directly consuming them, quality of the capsule would matter a lot for people’s health. By considering the demands of tablet production, this research gives a drug detection model on the basis of Convolution Neural Networks (CNN) which is one of the major algorithms in Deep Learning (DL). In recent years, DL computing paradigm was considered the best level among the Machine Learning (ML) section. Furthermore, in the ML field, it is the most extensively used computational method and has excellent outcomes on various large percipient tasks that match or exceed human performance. CNN algorithms help identify the type and name of the capsules. Images of diverse types of capsules are gathered for training the proposed CNN models DenseNet, Mobilenetv2 to detect and predict the name and information of capsule. This work investigates the DensNet architectures and MobileNetV2 in identifying the capsules and tablet classes. The Adaptive Moment Estimation (Adam) optimizer algorithm is proposed to train the model fast and to improve the convergence of the model. The predictive power of the model is measured using cross-entropy in the place of loss function. Finally, based on the best parameter for the dense model by cross-checking the results with various parameters, the detection method would choose. The results show an accuracy of almost 85% and above for all the models which are implemented in this research work. Therefore, these methods can be used for automating the process of identification and classification of the capsules.