Title: Artificial Intelligence Across Disciplines: Research, Innovation, and Intelligent Solutions
Editors: Dr. Subita Bhagat, Dr. A. Balamurugan, Dr. P. Krishna Kumar, and Mrs. S. Nandhini Devi
ISBN: 978-81-69857-83-3
Chapter: 8
DOI: https://doi.org/10.59646/815/08
Author: Dr. S. Diana Juliet
Abstract
Deep learning has fundamentally revolutionized computer vision, shifting automated visual inspection from brittle, handcrafted feature extractors toward resilient, hierarchical representation learning. This chapter examines the design, mathematical formulation, and operational deployment of an intelligent visual recognition framework applied to high-throughput automated optical inspection (AOI) in multi-layer semiconductor and printed circuit board (PCB) manufacturing. Modern industrial image streams present severe challenges, including micro-scale defect topologies, non-uniform specular reflections, dynamic background clutter, and severe class imbalance. To overcome these constraints, we formulate a multi-scale hybrid architecture integrating a Feature Pyramid Network (FPN) backbone with deformable convolutional layers and multi-head self-attention mechanisms, trained through focal loss and structural boundary regularization. Evaluated on an operational dataset of 50,000 ultra-high-resolution multi-spectral surface images, the proposed framework achieves a mean Average Precision (mAP@0.5) of 98.42%, an inference latency of 24.6 ms, and reduces false-alarm overkill rates by 82.7%.
Keywords: Deep Learning, Computer Vision, Visual Recognition, Intelligent Image Analytics, Deformable Convolution, Defect Detection