Informatics Publishing Limited

Vidyabharathi Dakshinamurthi ¹, Syed Ibad Ali ², T. Karthikeyan ³, Nanda Satish Kulkarni ⁴

Affiliations
1 Department of Computer Science and Engineering, Sona College of Technology, IN
2 School of Engineering, Ajeenkya DY Patil University, IN
3 Department of Information Technology, University of Technology and Applied Sciences - Salalah, OM
4 Department of Electronics and Telecommunication Engineering, Siddhant College of Engineering, IN

Abstract

Designing efficient mobile networks is crucial for meeting the growing demand for high-speed, reliable communication. However, existing convolutional neural network (CNN) architectures face challenges in capturing temporal dependencies, hindering their performance in mobile network design. The introduction highlights the increasing importance of mobile networks and identifies the limitations of current CNN architectures in capturing temporal dynamics. The problem statement emphasizes the need for an enhanced model that can effectively address temporal dependencies in mobile network design. This research addresses this problem by proposing a novel approach: Causal Convolution employing Almeida–Pineda Recurrent Backpropagation (CC-APRB). The causal convolution captures temporal dependencies by considering only past and present inputs, while the recurrent backpropagation optimizes the model parameters based on sequential data. The integration of these techniques aims to enhance the model ability to capture temporal features in mobile network data. The results indicate significant improvements in the performance of the CC-APRB model compared to traditional CNN architectures. The model demonstrates enhanced accuracy and efficiency in capturing temporal dependencies, making it well-suited for mobile network design applications.

Keywords

Causal Convolution, Almeida–Pineda Recurrent Backpropagation, Mobile Network Design, Temporal Dependencies, Deep Learning.

Full Text

     

Vidyabharathi Dakshinamurthi ¹, G. P. Suja ², P. Murugan ³, Sk. Riyaz Hussain ⁴

Affiliations
1 Department of Computer Science and Engineering, Sona College of Technology, IN
2 Research Center of Computer Science, Muslim Arts College, IN
3 Department of BBA, School of Management Studies, Vels Institute of Science Technology and Advanced Studies, IN
4 Department of Electronics and Communication Engineering, Rajiv Gandhi University of Knowledge Technologies, IN

Abstract

In recent years, image dehazing has gained significant attention in the field of computer vision and image processing due to its crucial role in enhancing visibility and improving visual perception. The presence of haze in images captured under adverse weather conditions or polluted environments poses a challenge to various computer vision applications, such as autonomous driving, surveillance, and satellite imagery. Traditional image dehazing methods often struggle to achieve optimal results, particularly in complex scenes with varying degrees of haze and intricate details. The need for a robust and efficient dehazing approach has become imperative for addressing real-world challenges in computer vision applications. Despite the advancements in traditional methods, a research gap exists in developing a comprehensive solution that can handle diverse atmospheric conditions and complex scenes effectively. The integration of deep learning techniques presents an opportunity to bridge this gap, leveraging the power of neural networks to learn and adapt to intricate patterns in hazy images. This research proposes a novel deep learning-based approach for image dehazing and visibility enhancement. A Convolutional Neural Network (CNN) architecture is designed to learn complex relationships between hazy and clear images, allowing the model to effectively remove haze and enhance visibility. The network is trained on a diverse dataset encompassing various atmospheric conditions and scene complexities to ensure generalization. Experimental results demonstrate the superior performance of the proposed deep learning approach compared to traditional methods. The model exhibits robustness in handling challenging scenarios, achieving significant improvements in image clarity, contrast, and overall visibility. The findings highlight the potential of deep learning in addressing the limitations of existing dehazing techniques.

Keywords

Deep Learning, Image Dehazing, Visibility Enhancement, Convolutional Neural Network, Computer Vision

Full Text