Exponentially-Krill Herd algorithm-based hybrid deep architecture for semantic image segmentation

View at publisher Summarise with VIRAT
Authors: S. Kanimozhi, K. Selvamani, Amol Dattatray Dhaygude, Vamsidhar Talasila
Cited by: 0
References: 52

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 paper proposes an effective semantic image segmentation model named Exponentially-Krill Herd algorithm based on Generative Adversarial Network + Deep Convolutional Neural Network (E-KH based GAN + DCNN). The optimization algorithm named Exponentially-Krill Herd algorithm (E-KH) is designed by the integration of Exponentially weighted moving average (EWMA) and Krill Herd (KH) algorithm, respectively. The segmentation results are generated using deep learning classifiers, specifically DCNN and GAN, in such a way that the segmented images can be fused based on the training error. Both the deep learning classifiers are allowed to segment the images simultaneously, such that bias-optimized process of the deep learning classifiers is accomplished using the proposed E-KH algorithm. The proposed method obtained better performance in terms of mean Intersection-Over-Union (mIOU), pixel accuracy, segmentation accuracy, and Jaccard coefficient of 75.17%, 85.05%, 84.46%, and 98.24% using the ADE20K database.

Fields of Study & Subjects

Advanced Neural Network Applications Medical Image Segmentation Techniques Metaheuristic Optimization Algorithms Research

Publication details

Journal / source
Statistics
Volume
60
Issue
5
Pages
1544–1578
Publication date
31 March 2026
Language
EN