基于图结构统计建模的多特征融合图像快速检索方法

陈奕孜; 赵丽娟

doi:10.19554/j.cnki.1001-3563.2026.17.027

PDF(1766 KB)

包装工程（技术栏目） ›› 2026, Vol. 47 ›› Issue (17) : 256-269. DOI: 10.19554/j.cnki.1001-3563.2026.17.027

自动化与智能化技术

基于图结构统计建模的多特征融合图像快速检索方法

陈奕孜¹, 赵丽娟^2,*

作者信息 +

Fast Image Retrieval Method with Multi-feature Fusion Based on Graph Structure Statistical Modeling

CHEN Yizi¹, ZHAO Lijuan^2,*

Author information +

文章历史 +

摘要

目的针对传统图像检索方法中单一尺度特征信息表达不充分、拓扑结构关联未被有效利用,导致检索准确率低、鲁棒性不足、精度与效率难以协同优化的问题,提出一种基于图结构统计建模的多特征融合图像快速检索方法。方法首先通过构建自适应跨层特征金字塔,引入层级权重学习机制与冗余过滤策略,自适应融合浅层细节与深层语义信息,增强模型对多尺度目标的判别能力。随后利用动态拓扑建模模块,将多尺度融合特征映射为图节点,结合特征相似度与空间约束动态优化边权,通过图卷积聚合局部-全局结构关联,提升特征鲁棒性。最后利用判别性主成分分析对高维特征进行压缩,优化特征维度,在保持检索精度的前提下进一步提升检索效率。同时建立多维度特征紧凑性统计评估体系,以类间和类内散度比、特征维度F-score、冗余相关系数及谱能量分布为量化指标,实现特征质量的全面评估引导调参优化效率。结果多数据集测试结果显示,所提方法可在12.3 ms内完成单张图像的特征提取与检索任务,在通用数据集上检索精度较当前方法提升3.5%~3.8%,在细粒度与工业专用数据集上提升4.2%~5.8%。结论所提方法使融合后特征的紧凑性与分离表征特性获得显著提升,能够有效兼顾复杂场景下检索任务的准确性、实时性与鲁棒性。

Abstract

The work aims to propose a fast multi-feature fusion image retrieval method based on graph structure statistical modeling to address the drawbacks of traditional image retrieval methods, including insufficient representation of single-scale feature information and ineffective utilization of topological structure correlations, which lead to low retrieval accuracy, poor robustness, and difficulty in collaborative optimization of accuracy and efficiency. Firstly, an adaptive cross-layer feature pyramid was constructed, with the hierarchical weight learning mechanism and the redundancy filtering strategy introduced, to adaptively fuse shallow detailed information and deep semantic information, enhancing the model's discriminative capability for multi-scale objects. Subsequently, a dynamic topological modeling module was designed, which mapped multi-scale fused features into graph nodes, dynamically optimized edge weights combining feature similarity and spatial constraints, and aggregated local-global structural correlations through graph convolution, improving feature robustness. Finally, Principal Component Analysis was utilized to compress high-dimensional features and optimize feature dimensionality, further boosting retrieval efficiency while maintaining satisfactory accuracy. Meanwhile, a multi-dimensional statistical evaluation system for feature compactness was established, with inter-class and intra-class divergence ratio, feature dimensionality F-score, redundancy correlation coefficient and spectral energy distribution as quantitative indicators, to realize comprehensive evaluation of feature quality and unsupervised early prediction of retrieval performance. Experimental results on multiple datasets showed that the proposed method could complete feature extraction and retrieval for a single image within 12.3 ms. Its retrieval accuracy outperformed the state-of-the-art methods by 3.5%-3.8% on general datasets, and 4.2%-5.8% on fine-grained and industrial dedicated datasets. The compactness and separable representation capability of the fused features are significantly improved by the proposed method, which can effectively balance the accuracy, real-time performance and robustness of retrieval tasks in complex scenes.

导出引用

陈奕孜, 赵丽娟. 基于图结构统计建模的多特征融合图像快速检索方法[J]. 包装工程. 2026, 47(17): 256-269 https://doi.org/10.19554/j.cnki.1001-3563.2026.17.027

CHEN Yizi, ZHAO Lijuan. Fast Image Retrieval Method with Multi-feature Fusion Based on Graph Structure Statistical Modeling[J]. Packaging Engineering. 2026, 47(17): 256-269 https://doi.org/10.19554/j.cnki.1001-3563.2026.17.027

中图分类号： TB11

参考文献

[1] 杨孟凡, 杨勇, 任鸽. 基于数据增强与多尺度的图文匹配优化方法[J]. 计算机工程与设计, 2025, 46(12): 3586-3592.
YANG M F, YANG Y, REN G.Optimization Method for Image-Text Matching Based on Data Augmentation and Multi-Scale Features[J]. Computer Engineering and Design, 2025, 46(12): 3586-3592.
[2] 张雪凝, 刘兴波, 宋井宽, 等. 面向大规模图像检索的哈希学习综述[J]. 软件学报, 2025, 36(1): 79-106.
ZHANG X N, LIU X B, SONG J K, et al.Survey on Hash Learning for Large-Scale Image Retrieval[J]. Journal of Software, 2025, 36(1): 79-106.
[3] 张小国, 李天宇, 史志豪, 等. 一种用于无人机景象匹配定位的异源图像快速检索方法[J]. 中国惯性技术学报, 2025, 33(10): 963-971.
ZHANG X G, LI T Y, SHI Z H, et al.A Fast Heterogeneous Image Retrieval Method for UAV Scene Matching and Positioning[J]. Journal of Chinese Inertial Technology, 2025, 33(10): 963-971.
[4] 张晓婷, 赵鹏宇, 潘如如, 等. 基于深度特征融合的格子织物图像检索方法[J]. 纺织学报, 2025, 46(8): 89-95.
ZHANG X T, ZHAO P Y, PAN R R, et al.Plaid Fabric Image Retrieval Method Based on Deep Feature Fusion[J]. Journal of Textile Research, 2025, 46(8): 89-95.
[5] 王智贤, 雷浩鹏, 王明文, 等. 一种面向手绘草图的多阶段跨域图像检索方法[J]. 计算机应用研究, 2025, 42(12): 3832-3840.
WANG Z X, LEI H P, WANG M W, et al.Multi-Stage Cross-Domain Image Retrieval Method Based on Hand-Drawn Sketches[J]. Application Research of Computers, 2025, 42(12): 3832-3840.
[6] 钟金彦, 陈俊, 李宇, 等. 基于MFF-SFE的遥感图文跨模态检索方法[J]. 中国科学院大学学报(中英文), 2025, 42(2): 236-247.
ZHONG J Y, CHEN J, LI Y, et al.Cross-Modal Retrieval Method Based on MFF-SFE for Remote Sensing Image-Text[J]. Journal of University of Chinese Academy of Sciences, 2025, 42(2): 236-247.
[7] 杨梦雅, 赵琰, 薛亮. 基于改进的Vision Transformer深度哈希图像检索[J]. 陕西科技大学学报, 2025, 43(4): 183-191.
YANG M Y, ZHAO Y, XUE L.Deep Hashing Method Based on Improved Vision Transformer[J]. Journal of Shaanxi University of Science & Technology, 2025, 43(4): 183-191.
[8] 刘洋, 毛克明. 基于自适应反馈机制的小差异化图像纹理特征信息数据检索[J]. 江苏大学学报(自然科学版), 2025, 46(1): 73-81.
LIU Y, MAO K M.Retrieval of Texture Feature Information Data for Small Differentiated Images Based on Adaptive Feedback Mechanism[J]. Journal of Jiangsu University (Natural Science Edition), 2025, 46(1): 73-81.
[9] 黄溪, 王先兵, 林海, 等. 基于自集成视觉Transformer的图像检索[J]. 武汉大学学报(工学版), 2024, 57(12): 1795-1802.
HUANG X, WANG X B, LIN H, et al.Image Retrieval Based on Self-Ensemble Vision Transformer[J]. Engineering Journal of Wuhan University, 2024, 57(12): 1795-1802.
[10] 尹晶晶, 潘丽丽, 王朝, 等. 基于双向语义嵌入的细粒度图文匹配方法[J]. 南京大学学报(自然科学), 2024, 60(5): 804-814.
YIN J J, PAN L L, WANG Z, et al.Bidirectional Semantic Embedding for Fine-Grained Image-Text Matching[J]. Journal of Nanjing University (Natural Sciences), 2024, 60(5): 804-814.
[11] 罗文培, 黄德根. 大模型增强的跨模态图文检索方法[J]. 小型微型计算机系统, 2025, 46(7): 1544-1553.
LUO W P, HUANG D G.Enhanced Cross-Modal Image-Text Retrieval with Large Models[J]. Journal of Chinese Computer Systems, 2025, 46(7): 1544-1553.
[12] 成晨, 曾子明, 孙守强. 基于DenseNet哈希的敦煌壁画多标签检索研究[J]. 数据分析与知识发现, 2025, 9(5): 146-156.
CHENG C, ZENG Z M, SUN S Q.Multi-Label Retrieval for Dunhuang Murals Based on DenseNet Hashing[J]. Data Analysis and Knowledge Discovery, 2025, 9(5): 146-156.
[13] 盖枚岭, 张辉辉, 秦琦冰. 基于特征金字塔网络的余弦四元组哈希图像检索方法[J]. 计算机工程与设计, 2024, 45(7): 2127-2133.
GAI M L, ZHANG H H, QIN Q B.Deep Cosine Quadruplet Hashing Based on Feature Pyramid Network for Image Retrieval[J]. Computer Engineering and Design, 2024, 45(7): 2127-2133.
[14] 侯青峰, 卢俊, 郭海涛, 等. 几何关系约束条件下的跨视角图像检索定位[J]. 地球信息科学学报, 2025, 27(10): 2316-2331.
HOU Q F, LU J, GUO H T, et al.Cross-View Image Retrieval and Localization under Geometric Relation Constraints[J]. Journal of Geo-Information Science, 2025, 27(10): 2316-2331.
[15] 张艳, 申多, 李增辉, 等. 基于特征分块与域间融合的跨域足迹图像检索方法[J]. 计算机辅助设计与图形学学报, 2025, 37(2): 303-312.
ZHANG Y, SHEN D, LI Z H, et al.Cross-Domain Footprint Image Retrieval Based on Feature Chunking and Domain Fusion[J]. Journal of Computer-Aided Design & Computer Graphics, 2025, 37(2): 303-312.
[16] 李景文, 龚旭, 蔡耀君, 等. 融合视觉特征与深层特征的遥感图像检索方法[J]. 计算机应用与软件, 2025, 42(12): 220-227.
LI J W, GONG X, CAI Y J, et al.A Remote Sensing Image Retrieval Method by Fusing Visual Features with Deep Features[J]. Computer Applications and Software, 2025, 42(12): 220-227.
[17] 丁功鸿, 黄山. 结合多特征与线性判别分析的图像检索[J]. 计算机应用与软件, 2024, 41(4): 212-218.
DING G H, HUANG S.Image Retrieval Based on Multipile Features and Linear Discriminant Analysis[J]. Computer Applications and Software, 2024, 41(4): 212-218.
[18] 赵圆圆, 李月军, 李昌庆, 等. 基于深度神经网络的船舶图像识别检索研究[J]. 舰船科学技术, 2024, 46(3): 174-177.
ZHAO Y Y, LI Y J, LI C Q, et al.Research on Ship Image Recognition and Retrieval Based on Deep Neural Network[J]. Ship Science and Technology, 2024, 46(3): 174-177.
[19] 刘华咏, 黄聪, 金汉均. 注意力增强的视觉Transformer图像检索算法[J]. 电子测量技术, 2023, 46(23): 50-55.
LIU H Y, HUANG C, JIN H J.Image Retrieval Method with Attention-Enhanced Visual Transformer[J]. Electronic Measurement Technology, 2023, 46(23): 50-55.
[20] 焦世超, 关日鹏, 况立群, 等. 深度语义关联学习的基于图像视觉数据跨域检索[J]. 计算机工程, 2024, 50(5): 190-199.
JIAO S C, GUAN R P, KUANG L Q, et al.Image-Based Cross-Domain Visual-Data Retrieval with Deep Semantic Correlation Learning[J]. Computer Engineering, 2024, 50(5): 190-199.
[21] LIU Y, PENG Y N, LI D X, et al.Crime Scene Investigation Image Retrieval with Fusion CNN Features Based on Transfer Learning[C]// Proceedings of the 3rd International Conference on Multimedia and Image Processing. Guiyang: ACM, 2018.
[22] HU J, SHEN L, SUN G.Squeeze-and-Excitation Networks[C]// Proceedings of the 2018 IEEE/CVF Conference on Computer Vision and Pattern Recognition. Salt Lake City: IEEE, 2018.
[23] LIN T Y, DOLLAR P, GIRSHICK R, et al.Feature Pyramid Networks for Object Detection[C]// Proceedings of the 2017 IEEE Conference on Computer Vision and Pattern Recognition (CVPR). Honolulu: IEEE, 2017.
[24] WANG C Y, ZHONG C M.Adaptive Feature Pyramid Networks for Object Detection[J]. IEEE Access, 2021, 9: 107024-107032.
[25] LI X.Graph-Based Superpixel Representation for Image Classification and Retrieval[C]// ICASSP 2019-2019 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP). [s. l.]: IEEE, 2019.
[26] SUCHARITHA G, KALYANI B J D, RAJU A S N, et al. Mutual Contextual Relation-Guided Dynamic Graph Networks for Cross-Modal Image-Text Retrieval[J]. Scientific Reports, 2025, 15: 34002.
[27] 彭永鑫. 面向图像检索的混合学习索引方法[J]. 计算机技术与发展, 2026, 36(2): 16-21.
PENG Y X.A Blended Learned Index Approach for Image Retrieval[J]. Computer Technology and Development, 2026, 36(2): 16-21.
[28] YANG J.Discriminative Feature Learning with Maximal Inter-Class and Minimal Intra-Class Scatter for Image Retrieval[J]. IEEE Transactions on Image Processing, 2018, 27(12): 5896-5908.
[29] 陈捷丰, 姚金良. 基于自相似嵌入和全局特征重排序的图像检索方法[J]. 浙江大学学报(工学版), 2025, 59(6): 1130-1139.
CHEN J F, YAO J L.Image Retrieval Method Based on Self-Similar Embedding and Global Feature Reranking[J]. Journal of Zhejiang University (Engineering Science), 2025, 59(6): 1130-1139.
[30] CHHABRA P, GARG N K, KUMAR M.Content-Based Image Retrieval System Using ORB and SIFT Features[J]. Neural Computing and Applications, 2020, 32(7): 2725-2733.
[31] NG H F, LIN C Y, MUINDISI T.Fast Image Matching Using Multi-Level Texture Descriptor[C]// Signal and Information Processing Association Annual Summit and Conference (APSIPA), 2014 Asia-Pacific. Siem Reap: IEEE, 2014.
[32] SUN X, JIN L, LI Z C.Attention-Aware Feature Pyramid Ordinal Hashing for Image Retrieval[C]// Proceedings of the 1st ACM International Conference on Multimedia in Asia. [s. l.]: ACM, 2020.
[33] WANG C, ZHAO J X, LI L L, et al.Automatic Graph Topology-Aware Transformer[J]. IEEE Transactions on Neural Networks and Learning Systems, 2025, 36(5): 8470-8484.
[34] HAN J, JIN X, SUN Y, et al.Adaptive Graph Convolution with Dynamic Edge Weights for Visual Retrieval[C]// Workshop on Representation Learning on Graphs and Manifolds, International Conference on Learning Representations. 2024.
[35] LI C S, LI T Z, HU X B, et al.DVHGNN: Multi-Scale Dilated Vision HGNN for Efficient Vision Recognition[C]// Proceedings of the 2025 IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR). Nashville: IEEE, 2025.
[36] 刘冉, 吴杰, 张蒙蒙. 人工智能与复杂系统的拓扑理论[J]. 数学年刊A辑, 2025, 46(3): 225-246.
LIU R, WU J, ZHANG M M.Topology in Artificial Intelligence and Complex Systems[J]. Chinese Annals of Mathematics, Series A, 2025, 46(3): 225-246.
[37] 王杰, 孟华. 基于点云整体拓扑结构的图像分类算法[J]. 计算机应用, 2024, 44(4): 1107-1113.
WANG J, MENG H.Image Classification Algorithm Based on Overall Topological Structure of Point Cloud[J]. Journal of Computer Applications, 2024, 44(4): 1107-1113.
[38] 杜先君, 丁家俊, 董明月. 拓扑数据分析在晶圆图缺陷模式分类中的高效应用[J]. 电子测量与仪器学报, 2025, 39(10): 185-196.
DU X J, DING J J, DONG M Y.Efficient Application of Topological Data Analysis in Wafer Map Defect Pattern Classification[J]. Journal of Electronic Measurement and Instrumentation, 2025, 39(10): 185-196.