Deep Learning Model for Ayurveda Doshas Classification Using Nail Image

Abstract

Analysis of nails is noteworthy from both biological as well as pathological point of view. Coming to Ayurveda, the nails act as an important indicator of impending bodily maladies. An exhaustive and conclusive evaluation leads to an effective course of action. Nail samples were gathered from individuals aged 17-30 from two local institutes in India. During the pre-processing stage, many techniques are employed to enhance image quality and reduce noise. These methods comprise rescaling, normalization, and segmentation techniques such as contrast-limited adaptive histogram equalization and histogram equalization followed by implementation of augmentation approaches. The advancements in deep learning have enhanced the convenience and precision of image processing. Classification strategy utilizing transfer learning has been suggested for detecting four groups of nail images based on ayurvedic principles. The approach incorporates a depth-wise separable convolutional neural network to classify the tri-dosha exhibits in human body. Furthermore, five state-of-the-art CNNs have been specifically designed for analyzing nail images, and their effectiveness has been assessed. The suggested model exhibits superior performance compared to the other high-end CNN architectures. Based on experimental research, the suggested model demonstrates a validation accuracy of 94.75% and a testing accuracy of 93.79% in the classification of nail images.