Medicinal plants have played a vital role in healthcare, especially in developing regions, where they are vital for healthcare. Traditionally, identification of these plants has been a slow and error-prone task, relying heavily on expert knowledge. This paper aims to automate the process of identifying medicinal leaves, focusing on Indian species known for their health benefits. We tested four convolutional neural network (CNN) models—VGG16, MobileNet, InceptionResNetV2, and our own PernNet model—to classify images of Indian medicinal leaves. The dataset included both healthy and diseased leaves in varying conditions. Using data augmentation and transfer learning, each model was fine-tuned to improve performance. PernNet specifically designed for leaf identification, outperformed the others with a 96.54% accuracy on test data. It has been compared with various models and then we concluded with the best performance. As “pern” in Sanskrit means leaf so we kept this name for our model as PernNet. Its success can be attributed to advanced features like compound scaling and multiscale feature extraction, which improve accuracy, stability, and robustness, particularly in handling variations in leaf texture and shape. PernNet balances accuracy and computational efficiency; hence, the tool has shown promise for use in real-time medicinal leaf identification, representing considerable advancement in automatic plant recognition.