Presentation Information
[18a-PA2-1]Object Detection Model Using Hybrid Quantum Classical Machine Learning
〇Ang Li1, Fei Bao1, Rei Sato2, Genki Okano2 (1.MACNICA, Inc., 2.Classiq Technologies G.K.)
Keywords:
Quantum Machine Learning,Quantum Computing,Quantum AI
In the field of image recognition, Convolutional Neural Network (CNN) and Vision Transformer (ViT) have become standard technologies. However, the feature representation capabilities of these classical methods are constrained by the classical computational framework, limiting their ability to extract complex correlation patterns. This study proposes a quantum-classical hybrid object detection model that leverages the high expressibility of quantum circuits to capture high-order correlations that are difficult for classical models to extract. Specifically, we constructed a "Hybrid Feature Enhancement Module" applied to feature maps extracted from ResNet18. This module integrates global context captured by a classical Transformer with local pixel correlations represented as quantum entanglement within a Variational Quantum Circuit (VQC). Furthermore, to stabilize training and mitigate the effects of noise and computational cost associated with quantum gradients, we introduced an "alternating fine-tuning" strategy that updates classical and quantum parameters in turn. Evaluation experiments using a traffic sign detection dataset demonstrated that the proposed quantum feature enhancement contributes to improved detection accuracy compared to a baseline model lacking the quantum module. We report these findings along with detailed results comparing multiple quantum circuit configurations (Ansatz).