Abstract:
This thesis explores the identification of medicinal plants in Bangladesh through images of plant
leaves using deep transfer learning. The primary aim of the study was to find the best and less
time-consuming procedure, which could enable people to identify different types of medicinal
plants by their leaves. To this end, we introduced and compared multiple deep learning models,
which are MobileNet V2, Inception V3, ResNet50, VGG16 & VGG19. Based on their
performance in the given tests, it was revealed that the MobileNet V2 had the best performance
among all the four models. It obtained an average accuracy of 94.54% on the dataset, which is
rather high. Regarding the performance characteristics, accuracy, recall rate, and F1-score are 95%
each. Combining these results shows how the model is great at identifying medicinal plants from
the leaves images. The research findings of this study have major managerial implications.
Regarding this model, the following advantages can be provided: The possibilities for its
application result in easier identification of medicinal plants by individuals, especially those from
rural and other underserved regions, thereby increasing the use of traditional medicine. This tool
may also be useful in an educational context since the nature of its approach is explanatory:
students and researchers may use this tool to get more information on the diverse variety of
medicinal plants found in Bangladesh. The model enhances environmental sustainability by
supporting the accurate identification and that causes preservation of plant species. By reducing
the risk of misidentification and over-harvesting, it plays a great role in the conservation of
biological diversity. The research we presented highlights the potential of deep learning
technologies, like MobileNet V2, in the prospects of effective relations between humans and the
environment. Besides establishing the foundation for future research, this work also creates the
roadmap for the creation of helpful and easy to use tools that may allow individuals to apply
modern technologies to knowledge obtained from more conventional sources.
