Improved Medicinal Plant Leaf Classification using Transfer Learning and Model Averaging Ensemble of Deep Convolutional Neural Networks

Identification of medicinal plants is a difficult undertaking. It needs an expert’s presence to recognize them. The goal of this study is to enhance existing classification and recognition accuracies of medicinal plant leaf. A model averaging ensemble of five (5) convolutional neural networks namely: RestNet50, RestNet101, ResNet152, VGG16 and VGG19 was developed which resulted in 98.22% overall accuracy, 98.40% precision rate, recall of 98.13% and 98.25% F1-score. Furthermore, transfer learning was used in the training of each network. This is done to lessen the training period of each CNN. Additionally, actual image count was increased to add variance to the data and augment training accuracy. According to the study’s findings, categorizing and identifying medicinal plants using model averaging ensemble is a viable attempt in boosting overall recognition accuracy rate.