基于并行权重自适应k-邻域算法的图像分类方法  

Image classification method based on parallel weight adaptive k-neighborhood algorithm

作　　者：苗水清 闫文耀[1] 吴梦蝶 MIAO Shuiqing;YAN Wenyao;WU Mengdie(Department of Data Science and Engineering,Xi'an Innovation College of Yanan University,Xi'an,Shaanxi 710100,China)

机构地区：[1]延安大学西安创新学院数据科学与工程学院,陕西西安710100

出　　处：《贵州师范大学学报（自然科学版）》2023年第2期113-120,共8页Journal of Guizhou Normal University：Natural Sciences

基　　金：陕西省教育厅科研计划资助项目(21JK0997)。

摘　　要：针对现有KNN算法识别率低的问题,提出了一种并行权重自适应k-邻域算法。该方法首先结合多线程技术,并采用分类组合的多个KNN单元进行识别以提高执行效率;其次在分类组合KNN算法中采用深度学习模型对各个类别进行了系数权重自适应设定,进而降低传统KNN和分类组合KNN,由于单纯类别个数的多少进行决策或者通过人为设定类别比例进行决策而引起的分类误差。通过在Fashion MNIST手写数据集进行实验,结果表明:该算法将传统的KNN算法分类正确率提高到97%左右,对实际应用具有一定的价值。Aiming at the problem of low recognition rate of existing KNN algorithms,a parallel weight adaptive k-neighborhood algorithm is proposed.Firstly,the method combines multithreading technology and uses multiple KNN units combined by classification to improve the execution efficiency;Secondly,in the classification and combination KNN algorithm,the deep learning model is used to adaptively set the coefficient weight of each category,so as to reduce the classification error caused by the traditional KNN and classification and combination KNN because of the number of simple categories or by artificially setting the category proportion.Through the experiment on Fashion MNIST handwritten data set,the results show that the algorithm improves the classification accuracy of the traditional KNN algorithm to about 97%,and has certain value for practical application.

关 键 词：并行权重自适应 k-邻域算法 深度学习 KNN 图像分类 

分 类 号：TP391.9[自动化与计算机技术—计算机应用技术]