A Novel Approach for Authenticating Churna Medicine Using Transfer Learning

In the field of Siddha medication system, Churna is a fine herbal powder used in Indian medicine. Identifying the specific type of Indian siddha medicine (churna) based on visual characteristics is a challenging task. This paper proposes an efficient method for churna identification using transfer learning techniques A manual dataset of 1555 churna images was created in which 1120 images are given for training, 195 images are given for validation and 240 images are given testing purposes. Experimentation has been carried out using transfer learning techniques such as VGG16, ResNet50, DenseNet121, MobileNetV2, and ResNet101. The experimental analysis shows that ResNet101 achieved the highest accuracy of 84% compared to the other pre trained models. This approach offers a promising solution for accurate churna identification even for the individuals who does not have a prior knowledge about Churna.