用原子量子计算机实现图像匹配，仅需不到24个原子即可识别工业物体。
Rydberg Vision via frugal Quantum Image Fingerprinting
- 通过边缘提取与简化算法将图像转为稀疏点云，减少原子数量
- 利用里德堡相互作用编码图像几何，生成固定长度的量子指纹向量
- 首次在模拟量子计算中用静态结构因子做图像检索，适合小样本学习
基于门模型的量子图像处理受限于量子比特稀缺和量子态制备的高开销，难以应用于真实几何数据。本文提出一种面向中性原子模拟量子计算机的原生量子图像匹配框架，改进了先前的稀疏点表示（SDR）方法。经典预处理流程——使用Sobel边缘检测结合拉默-道格拉斯-普克（RDP）算法——将输入图像转换为几何忠实的稀疏点云，原子数显著减少。该原子布局通过QuEra Aquila设备的Bloqade SDK虚拟嵌入可编程光镊阵列中，图像几何信息由里德堡哈密顿量中的距离依赖范德华相互作用物理编码。经过时间演化后，采用两个可观测量提取多体指纹：经过皮尔逊归一化的双站点关联矩阵，捕捉阻塞诱导的相关结构；以及在固定波矢网格上计算的二维静态结构因子，生成长度恒定的指纹向量。第一阶段通过余弦相似度对指纹向量进行图像匹配，适用于傅里叶域描述子。第二阶段扩展至量子储备池计算（QRC），实现极低训练数据与训练轮次下的机器学习，作为初步概念验证。使用Bloqade软件栈的仿真表明，工业物体的匹配成功，通常仅需少于24个原子。据我们所知，这是首个将静态结构因子——凝聚态物理中的量子可观测量——作为图像检索描述子应用于模拟量子计算的案例。
原文摘要 · Abstract (English)
Gate-based quantum image processing is constrained by qubit scarcity and the high overhead of quantum state preparation, limiting its applicability to realistic geometric data. We introduce a quantum-native framework for image matching on neutral-atom analog quantum computers that advances our earlier Sparse-Dots Representation (SDR) approach. A classical pre-processing pipeline -- Sobel edge extraction followed by the Ramer--Douglas--Peucker (RDP) algorithm -- converts an input image into a geometrically faithful Sparse-Dots point cloud of substantially fewer atoms. This atom layout is virtually embedded into the programmable tweezer array of QuEra's Aquila device via its Bloqade SDK, where the image geometry is encoded physically in the distance-dependent van der Waals interaction term of the Rydberg Hamiltonian. After time-evolution, we extract the many-body fingerprint of each image using two observables -- the Pearson-normalized two-site correlation matrix which encodes the blockade-induced correlation structure of the quantum state, and the two-dimensional static structure factor evaluated on a fixed wavevector grid, yielding a fingerprint vector of constant length regardless of atom count. In Stage~1, image matching is performed by cosine similarity on the fingerprint vectors, a scale-invariant metric appropriate for Fourier-domain descriptors. In Stage~2, this approach is extended to quantum reservoir computing~(QRC) to enable machine learning via dramatically reduced training data and training cycles, as a preliminary proof-of-concept. Simulations using the Bloqade software stack confirm successful matching of industrial objects, often with fewer than 24 atoms. To our knowledge, this constitutes the first application of the static structure factor -- a condensed-matter quantum observable -- as an image retrieval descriptor in an analog quantum computing context.
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。