Premium
Dorsal Hand Vein Recognition Using Very Deep Learning
Author(s) -
Kumar Rajendra,
Singh Ram C.,
Kant Shri
Publication year - 2021
Publication title -
macromolecular symposia
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 0.257
H-Index - 76
eISSN - 1521-3900
pISSN - 1022-1360
DOI - 10.1002/masy.202000244
Subject(s) - deep learning , computer science , artificial intelligence , pooling , pattern recognition (psychology) , dorsum , convolutional neural network , normalization (sociology) , binary classification , artificial neural network , deep neural networks , anatomy , support vector machine , medicine , sociology , anthropology
Designing a deep neural network for multiclass classification is a challenging task in comparison to binary classification. In this paper, a very deep learning model for dorsal hand vein recognition is presented. Deep learning models may contain hundreds of hidden layers. The increase in layers makes a model complex and it takes much time to train especially on CPU instead of GPU. The proposed model consists of network of Convolution layers and ReLU layers followed by one Global Average Pooling layer and two dense layers. To keep the model stable, batch normalization is applied on input and hidden layers and Adam optimizer is used to optimize the model. For the training and testing purpose two self‐constructed datasets are used, first having 1200 dorsal hand vein images of children of age ranging from 5 to 12 years and second having 2000 dorsal hand vein images from 100 adults. The accuracy of the proposed model is compared with two recent contributions, which also used self‐constructed datasets. The accuracy of the proposed system observed is 98.33% when trained and tested with children' dataset, and 100% when trained and tested with adults' dataset.