Hybrid Convolutional Neural Network Model for Robust and Sustainable Underwater Image Enhancement

View at publisher Summarise with VIRAT
Authors: Mr. Manikandan B, M. Ramkumar Prabhu, G. Charulatha, Aarthi Sivasankaran, V. Suresh Kumar, K. Sathish
Cited by: 0
References: 15

Summary

No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.

Summarise with VIRAT Free to try with your Science Catalogs account.

Abstract

This work presents a comparative deep learning architecture utilising Xception, VGG16, and a Combined CNN model to enhance and assess underwater photos. The proposed study aims to improve SDGs 9 (Industry, Innovation, and Infrastructure) and 14 (Life Below Water) by improving the quality of images, the accuracy of feature reconstruction, and the accuracy of categorisation. The investigations show that Xception has great structural strength and noise resistance, with a PSNR of 39.60 dB and an SSIM of 0.985. VGG16 has a low log loss of$\mathbf{0. 1 5 2 5}$, which means that its predictions are more accurate and dependable. The Combined Model, which is a mix of feature fusion and generalisation, gets perfect accuracy (1.0000) in simulated multi-class evaluation, even though it has a higher log loss and a much lower PSNR. Consistent measurements of entropy, contrast, and brightness across all models show that luminance is stable and the image is dynamic. The results show that Xception is better at fine-tuning minor details than VGG16, but VGG16 is the best choice for balancing processing speed and perceptual quality. This study integrates breakthroughs in deep learning with global goals for sustainable technology development to demonstrate the efficacy of deep convolutional architectures in sustainable picture enhancement, marine monitoring, and intelligent vision systems.

Fields of Study & Subjects

Image Enhancement Techniques Advanced Neural Network Applications Advanced Image Processing Techniques

Publication details

Journal / source
International Conference on Communication, Computing and Electronics Systems (ICCCES)
Pages
1080–1085
Publication date
21 January 2026
Language
EN