基于视觉描述符的图像大数据分类算法仿真  

Simulation of Image Large Data Classification Algorithm Based on Visual Descriptor

作　　者：曹敏 曹东朗 CAO Min;CAO Dong-lang(Shanxi Police College,Taiyuan Shanxi 030401,China;School of Modern Educational Technology,Shanxi University,Taiyuan Shanxi 030401,China)

机构地区：[1]山西警察学院,山西太原030401 [2]山西大学现代教育技术学院,山西太原030401

出　　处：《计算机仿真》2024年第4期170-174,300,共6页Computer Simulation

基　　金：山西省哲学社会科学规划办公室:大数据融合创新视角下山西战略性新兴产业链集群发展现状及潜力研究(2022YD165);山西省教育厅:基于多维驱动的信息安全专业人才培养机制研究(J20221297);山西警察学院:公安院校的“金课”建设研究(YJ202126)。

摘　　要：图像大数据化是不可阻挡的科技进程,但随着图像数量的增多,传统分类算法在图像识别与分类上具有一定的局限性。为解决大数据图像分类的精确度低下的问题,提出一种融合图像视觉描述符与图像初级特征的分类算法。首先利用迁移学习的优势,从VGG18的最大池化层提取图像的初级特征;然后加个图像预处理,采用“82圆型LBP算子”与“化Canny算子”分别提取同质纹理描述符与边缘直方描述符;最后将图像基础特征与视觉描述符相融合构建基于支持向量机的图像识别分类模型(DES-SVM)。仿真结果表明,经图像视觉描述符与图像初级特征相融合的建模方式,有效的提高了图像分类的精确度,较传统SVM模型相比,DES-SVM模型在UKB图像库与ZBD图像库上准确率、召回率与F指标分别提高了7.85%、8.42%和8.13%。构建的DES-SVM图像识别分类模型通过视觉描述符提取的方式有效的提升了模型的性能。Image big data is an irresistible scientific and technological process,but with the increase of the number of images,traditional classification algorithms have certain limitations in image recognition and classification.In order to solve the problem of low accuracy of large data image classification,this paper proposes a classification algorithm that integrates image visual descriptors and image primary features.Firstly,the primary features of the image were extracted from the maximum pooling layer of VGG18 by using the advantages of transfer learning,and then an image preprocessing was added,and the homogeneous texture descriptor and the edge histogram descriptor were extracted by using the"82 circular LBP operator"and the"Canny operator"respectively.Finally,an image recognition and classification model based on support vector machine(DES-SVM)was constructed by fusing basic image features and visual descriptors.Simulation results show that the proposed method can effectively improve the accuracy of image classification.Compared with the traditional SVM model,the accuracy,recall and F index of the DES-SVM model on the UKB image database and ZBD image database are increased by 7.85%,8.42%and 8.13%respectively.The DES-SVM image recognition and classification model constructed in this paper effectively improves the performance of the model through the extraction of visual descriptors.

关 键 词：特征融合 图像分类 大数据 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]