Improvement of Smoker Prediction System Based on Hyperspectral Image with Hybrid Deep Learning Model

The tongue image classification system has been widely used in medical interests and health diagnoses. This research emphasizes improving classification accuracy performance in the Smoker prediction system based on the location analysis of the smoker melanosis distribution on the tongue image. The tongue diagonalization technique developed is a non-invasive method based on hyperspectral imaging (HSI). Various considerations and In-depth architecture learning have been proposed to overcome the analysis of HSI data and have obtained relatively high classification completion. This study uses the Convolutional Neural Network (CNN) architecture in the spectral-spatial configuration used for feature extraction and classification. CNN to do some testing. Researchers classified it as Single CNN and Hybrid-CNN. In the Single CNN algorithm, there are two architectures created, namely CNN-Autoencoder and CNN-Alex net. In the Hybrid-CNN algorithm, two architectures are designed, namely Proposed Hybrid-CNN with one branch and Hybrid-CNN-Resnet18 with eight branches. Learn more about the kernel in each different subject segmentation and look at the kernel classification. Therefore, the Hybrid-CNN model is proposed to be able to make hybrid architecture and hybrid convolution scale. The approved Proposed Hybrid-CNN model, supported about Lateral A can reach 90,60%, Lateral B reaches 86,5%, and Doctor Perception reaches 99,2%. In the CNN-Resnet18 Hybrid model obtained, Lateral A can reach 89,4%, Lateral B gets 84,6%, and Doctor Perception reaches 97,4%. In general, the results of the completion of the approved model have achieved better performance.