Title: AI, Innovation and Global Transformation: Interdisciplinary Perspectives on Technology, Business and Society
Editors: Dr. J. Preetha, and Dr. Siddhartha Mehrotra
ISBN: 978-81-69857-64-2
Chapter: 28
DOI: https://doi.org/10.59646/809/28
Author: E. Priya
Abstract
Modern computer vision has transitioned from manual descriptor engineering toward deep hierarchical representation learning, fundamentally altering automated visual analytics across healthcare, autonomous navigation, industrial quality control, and surveillance systems. While standard convolutional and self-attention vision networks achieve remarkable diagnostic and predictive accuracy on benchmark image distributions, deploying these systems into real-world automated visual environments poses severe computational bottlenecks, spatial sensitivity failures, and domain drift vulnerabilities. This chapter presents an end-to-end framework combining deformable dynamic convolutions, multi-scale shifted window self-attention, and contrastive latent self-calibration for intelligent image analytics. Through structured multi-stage feature extraction, the proposed architecture simultaneously captures fine-grained spatial edge topologies and long-range semantic dependencies under severe atmospheric degradations and real-time edge hardware constraints. Systematic empirical evaluations across multi-class benchmark datasets establish performance improvements in mean Average Precision, structural segmentation fidelity, and computational inference efficiency, charting a robust pathway for next-generation automated image understanding engines.
Keywords: Computer Vision; Deep Learning; Image Analytics; Vision Transformers; Deformable Convolution; Multi-Scale Feature Fusion.