Residual Architecture for Image Recognition

Residual Architecture for Image Recognition

Author Name : Manikandan P, Dr. Lekha J, Prakash R, Praveen Kumar S

ABSTRACT

The tongue plays a significant role in the human body. The tongue is an organ which is connected to other important parts of the human body. If any parts get affected, tongue has its unique characteristic to reflect the change in it. So, one has the leverage to find the disease in initial stage by diagnosing the tongue. Ancient Traditional Chinese medicine(TCM) followed this method to predict the disease. In this paper, a prediction system to diagnose diabetes mellitus using tongue image is proposed. The system predicts the diabetes based on the features of the tongue like the color distribution and texture etc.., The system extracts features from the images and stores it. By using the features obtained and with the help of decision tree classifiers, we can predict the results. In this system, we additionally implement the RESNET architecture over the other existing architectures. The existing architecture has its disadvantages in its accuracy while adding additional layers to it. Layers are increased to improve the efficiency of the system. It also suffers with two major problems (vanishing gradient and degradation.). As a result, the existing system suffers in its increased error rate and accuracy. Using RESNET architecture gives more consistency to the system. It overcomes the vanishing gradient and degradation by using a term called skip connection. By using skip connection, it can add upto 100 layers easily while others are struggling to achieve more than 20 layers. In addition to it, existing system uses additional physical device named SEM(Sequential Electron Microscope) camera to capture images. Here we are proposing mobile cameras as the potential of mobile cameras got increased.

Keyword: CNN, Neural networks, RESNET, Vanishing gradient