Automated Predictive Modeling for Identifying Medicinal Plant Parts using CNNs

The ResNet50 architecture, a deep convolutional neural network (CNN), is used for accurate image classification tasks. Training and evaluation are conducted on a data set comprising 10 species of medicinal plants, with three categories: leaves, roots, and flowers per species. Transfer learning is utilized with pre-trained ImageNet weights, and the classification layers are modified to align with the dataset. Data augmentation and fine-tuning techniques are applied to enhance model performance, resulting in improved prediction accuracy. The experimental results indicate that ResNet50 achieves a training accuracy of 97.79%, a validation accuracy of 89.24%, and a test accuracy of 88.79%. The proposed approach automates the identification and prediction of medicinal plant parts, offering potential applications in herbal medicine, scientific research, and conservation of biodiversity.