Deployment Strategies for AI-based Computer Vision in Industrial Quality Control

The integration of Artificial Intelligence (AI) and Computer Vision (CV) is revolutionizing industrial quality control by enabling automated and reliable defect detection. While deep learning models achieve high accuracy on benchmark datasets such as MVTec-AD, their effective deployment in real manufacturing environments remains challenging due to data imbalance, limited availability of realistic defect samples, and heterogeneous computing infrastructures. This paper investigates deployment strategies for AI-based CV systems across edge and cloud platforms, analyzing the trade-offs between accuracy, latency, and computational cost. Two representative convolutional neural network architectures, ResNet18 and EfficientNet-B0, are systematically evaluated using advanced data augmentation and balancing techniques. Experimental results demonstrate that EfficientNet-B0 achieves superior classification performance, with accuracy up to 86.22%, while edge GPU deployment ensures the lowest inference latency (120ms) suitable for real-Time industrial inspection, albeit requiring investment in local infrastructure expansion. Conversely, cloud deployment offers scalability and reduced capital expenditure but incurs latency penalties (192-311ms) that limit time-critical applications. The findings highlight the importance of balancing performance requirements with infrastructure costs when designing deployment strategies for AI-driven quality inspection systems in modern manufacturing.