Abstract:
"Ayurveda is a traditional healthcare system in Sri Lanka that dates to more than 3000 years.
Ayurveda uses raw materials obtained from the nature to produce medicine that can be used to
treat various health conditions and diseases. Plant leaves are one of the main ingredients used to
prepare Ayurvedic medicine. Plant leaves used in the process of preparing Ayurvedic medicine is
normally collected from gardens and forests. Sometimes the individuals who collect hem may not
be able to recognize the Ayurvedic plants leaves correctly which could result in the use of incorrect
leaf types in the preparation of the medicine. This could cause the medicine to be ineffective or
even toxic in some conditions. Furthermore, Ayurvedic plants get destroyed unintentionally when
gardens and forests get cleared as many individuals are not able to recognize Ayurvedic medicinal
plants on their own this could lead to the endangerment and extinction of these valuable plants.
To solve this problem the author has implemented a system that is able to recognize Ayurvedic
medicinal plants using their leaves when in their natural environment. The author took two
approaches in implementing this system to find the best solution. The first method involved the
building of a Convolutional Neural Network architecture (CNN) from scratch the other method
involved the training of multiple pre-trained models using transfer learning and combining them
by using different ensemble techniques to find the best ensemble technique.
During testing, accuracy was used as the evaluation metric to determine the performance of the
two models. The CNN model gave an accuracy of 50% and the Ensemble model showed higher
performance with an accuracy of 95%."
