Quantum artificial vision for defect detection in manufacturing

TL;DR

This work investigates quantum machine learning for industrial defect detection by implementing QSVM on a gate-based quantum computer and QBoost on a quantum annealer, then benchmarking against classical methods on the GDXray Castings dataset. Using PCA for dimensionality reduction, both quantum approaches achieve higher precision and F1 scores than classical baselines, with QBoost particularly leveraging current quantum annealers to handle industrial-scale problems. The study notes that quantum training can be executed on quantum hardware while deployment occurs on classical systems, offering practical pathways for near-term industrial adoption and energy efficiency advantages. Future directions include benchmarking against CNN-based approaches and expanding quantum-inspired techniques for real production lines.

Abstract

In this paper we consider several algorithms for quantum computer vision using Noisy Intermediate-Scale Quantum (NISQ) devices, and benchmark them for a real problem against their classical counterparts. Specifically, we consider two approaches: a quantum Support Vector Machine (QSVM) on a universal gate-based quantum computer, and QBoost on a quantum annealer. The quantum vision systems are benchmarked for an unbalanced dataset of images where the aim is to detect defects in manufactured car pieces. We see that the quantum algorithms outperform their classical counterparts in several ways, with QBoost allowing for larger problems to be analyzed with present-day quantum annealers. Data preprocessing, including dimensionality reduction and contrast enhancement, is also discussed, as well as hyperparameter tuning in QBoost. To the best of our knowledge, this is the first implementation of quantum computer vision systems for a problem of industrial relevance in a manufacturing production line.

Figures (7)

• Figure 1: [Color online] Examples of images of the "Castings" group, with the corresponding bounding boxes surrounding the defects. Picture taken from Ref.gdxray, reproduced with permission.
• Figure 2: Example of the different contrast enhancement techniques used in this section, along with the transformation each of them produces on the image's histogram. Image from the Sci-kit image tutorial, see Ref.skimage.
• Figure 3: [Color online] F1 score results for different regularisation parameter values with and without RUS, with an initial number of 10 weak classifiers.
• Figure 4: [Color online] F1 score results for different regularisation parameter values with and without RUS, with an initial number of 50 weak classifiers.
• Figure 5: [Color online] Inference time for different regularisation parameter values, with an initial number of 10 weak classifiers and different tree depths.