Deep Convolution Neural Network Model for Indian Sign Language Classification

Abstract

Communication gap between non-hearing and hearing people results in many interaction difficulties across the globe. Indian sign language is the traditional communication alternative in our country. Recently, the integration of deep learning with convolution neural network plays a major role in solving various image classification problems. The proposed deep convolution neural network model is prepared with six convolution layers and three fully connected layers by altering different parameters. This model is evaluated based on 1000 number sign images. These datasets are created in college laboratory with 100 different students. Here, 1–10 number sign images are collected, and the proposed CNN model by six convolutional layers with 1000 epochs is applied. This study highlights the training and validation accuracy analysis as well as training and validation losses. Different performance metrics are calculated to find out the accuracy of the proposed model. The aim of developing the proposed model is to find out classification accuracy for each class. This is multiclass problems as each number signs contain 100 images. This model can predict 10 different signs classes. We have taken 200 signs from 1000 number signs for testing purpose. The results indicate that the accuracy of the proposed method is rapidly increased by increasing the epochs. We have achieved 73% average classification model accuracy.