A Comparative Study of Classical and Quantum Machine Learning Techniques for Industrial Defect Classification

This work presents a comparative evaluation of classical and quantum machine learning models applied to a real-world industrial classification task. The use case is provided by MBDA and involves classifying images of surface-mount technology solder joints into three defect severity classes. To investigate the viability and performance trade-offs of quantum models under realistic constraints, we followed two parallel processing strategies. The first employs conventional deep learning architectures, namely YOLOv8 and a custom Convolutional Neural Network (CNN), trained on augmented full-resolution images. The second strategy compresses image data via a domainspecific method, enabling compatibility with quantum hardware constraints. We trained and evaluated different quantum machine learning models, including a Quantum Support Vector Machine (QSVM), a Variational Quantum Classifier (VQC), and a Quantum Convolutional Neural Network (QCNN). Our results highlight the trade-offs between expressivity and feasibility across classical and quantum approaches and provide insights into the applicability of quantum models in industrial scenarios.