Recent years have seen a growth in the field of artificial intelligence (AI), with deep learning (DL) approaches offering up new opportunities for cutting-edge outcomes in an increasing number of fields. The use of technology in e-government applications to improve both the systems and citizen-government interactions is still hindered by a variety of challenges. The authors explore the issues with e-government systems in this chapter and offer a paradigm for automating and streamlining e-government services. Convolutional neural networks (CNNs) and other state-of-the-art techniques, such as transfer learning and deep ensemble learning, have been used to classify problems with high accuracy. Our overall objective is to use trustworthy AI methods to improve the current state of e-government services and lower processing times, costs, and citizen enjoyment. Several instances will also be included in the chapter to demonstrate how DL techniques can be applied in practical situations.
Top1. Introduction
Deep neural networks are widely employed in a variety of real-world applications, including computer vision, pattern recognition, and natural language processing. They are taught using tensor-based processes such as convolution neural networks and matrix multiplication and feature a layer-by-layer design. In supervised learning, a neural network is trained to achieve the maximum overall accuracy through a learning process using supplied training data. In such instances, the accuracies of the classes are frequently dissimilar. Some classes, in particular, may not be accurate enough, despite the fact that certain classes are more significant than others in certain applications for specific consumers. This issue can affect not just certain classes, but also certain types of data objects, such as implicit sub-classes. Fixing or tweaking a neural network to further increase the accuracy for certain classes or objects after training is not straightforward since the entire network has already been tuned via hundreds of thousands of iterations and there are intricate interconnections among the features and outputs. There are some applications used in security analysis.
Face detection and recognition is a fascinating and fast evolving research area with several applications. A significant number of face detection and recognition computations has been developed over time. Faces in images may be easily seen and distinguished by humans, but not by robots. Face detection and recognition may be accomplished in a variety of ways using machine learning. The multidimensional nature of the human face necessitates the adoption of a high-quality computer algorithm for recognition. To recognise faces in images, look for patterns such as height, skin colour, and the width of other elements of the face such as the lips, nose, and eyes. There is obviously a pattern, with varying dimensions for distinct faces and similar dimensions for related faces. We need to translate a certain face into numbers. Improving face recognition performance has been an on-going battle since the first algorithm was developed. Alex Pentland and Matthew Turk employed Principal Component Analysis (PCA) in 1991. Eigenface approaches are using different strategy for all modern face recognition systems (Mahammad et al., 2023).
In computer vision applications, there are two categories of difficulties: face detection and recognition. These techniques detect the presence of a human face in an image or video, whereas face recognition validates identity by using facial characteristics. As a result, face detection is a critical problem in 1st stage face recognition process. Because of the variety in human appearances, such as the presence of eyeglasses, the orientation of the face, the presence of facial hair, differences in lighting conditions, and picture quality, face identification is a difficult task for robots. Face identification is a more difficult procedure since it must account for inherent face traits such as age, occlusion, facial emotions, and so on.
Face detection and recognition are hybrid techniques combine the advantages of feature-based and holistic methods. The key challenge is the restricted number of outstanding features capable of properly extracting the vital information required for face detection. Machine learning approaches enable end-to-end systems to learn a large number of attributes required for optimal face detection and recognition tasks (Shukla et al., 2020).
Convolutional neural networks have been found to be the most useful and accurate type of deep learning technology for facial recognition. The neural networks are used to reduce dimensionality and classifier trained the present images in face detection and recognition applications. The object are identifying and recognizing after learning facial features during training time and encountered the problem by the system. The CNN models are occupying face detection and recognition solutions after inspired by the ImageNet Large Scale Visual Recognition Challenge (INLSVRC). Common and most recent CNN architecture, such as VGGnet, RESNET, and MobileNetV2, are also better examples of how to solve common and most recent CNN architecture.