الفهرس | Only 14 pages are availabe for public view |
Abstract The operator of a nuclear facility uses proprietary knowledge, facilities, and equipment to evaluate how mechanisms and systems respond to radiation and other environmental aspects. Areas of knowledge include radiation physics, reactor physics and engineering, experimental technology, instrumentation, and data acquisition. The Operator is also responsible for providing the organization with essential sustenance to support the security center by recognizing the individual workers in the nuclear facility. In order to avoid constraints, the multimodal biometric schemes provide more accuracy and security to the user’s data. In recent years, biometrics metrics related to the human biological and behavior features act as an imperious character in-person verification. The main objective of this research is to propose a new feature recognition hand vein model. The hand vein modalities have several benefits namely, the vein in hand is the internal features of the body, which cannot be faked. Moreover, recognition of the hand vein patterns approach has proposed different Convolutional Neural Network (CNN) models. This approach is routinely welllearned in a way to get features from the main pattern using Region of Interest (ROI). Though, the poor quality of the hand vein image still impose limitations to the extension of its usability. To begin with, by applying the method of Generative Adversarial Networks (GAN) data augmentation, the performance gets improved by adding truer images and applying ROI in a hand vein image feature extraction. The suggested approach has been tested on the datasets of hand veins to decrease the overfitting in the fully connecting layer of CNN which proves to be effective. In total, 1575 hand vein images from 100 subjects were applied to authorize the proposed approach for hand veins. A high accuracy (>99.8%) and low False Rejection Rate (FRR) (<0.99%) had been achieved by applying the suggested approach, when compared to the existing CNN classifiers. Also, face recognition of the masked individuals due to the spread of the COVID-I9 pandemic in early 2020 has been proposed. The World Human Organization (WHO) advised all people in the world to wear face-mask to limit the spread of COVID-19. Many facilities required that their employees wear face-mask. For the safety of the facility, it was mandatory to recognize the identity of the individual wearing the mask. Hence, face recognition of the masked individuals has been required. In this research, a technique has beenproposed based on a mobile-net and Haar-like algorithm for detecting and recognizing the masked face. Recognizing the authorized person that enters a nuclear facility in case of wearing mask could be achieved using a mobile - net. In addition, the Haar-like features are applied to detect the retina of the person and extract the bounding box around the retina which is compared with the dataset of the person without the mask for recognition. The results of the proposed model, which has been tested on a dataset from Kaggle, yielded 0.99 accuracies, a loss of 0.08 and F1. Score 0.98. Multi-modal systems have been proposed in which a single hand dataset has been used to get multiple traits for the evidence collected from multiple traits. Hand vein feature has been extracted and by extraction, a few features of hand geometry from the same dataset authorizations have been achieved with the help of fusion at score-level and feature level. Fusion at the feature level operates better because the informal implementation of these two traits are the most widely recognized biometrics in most requests. The use of 1D Res-net 50 for the Hand Geometry (HG) recognition model with an accuracy of 98.9% and EER of 0.027 was investigated. Different fusion levels have been investigated in this research. Score fusion level is based on arithmetic rule and weighted sum rule. The experiments showed that both rules have accuracy because they are rule-based techniques with an accuracy of 99.33% and ERR of 0.023. A weighted feature fusion model was proposed in this research and the experiments showed that feature fusion has a better accuracy (99.59%) than that of fusion at score level. |