当前位置: X-MOL 学术Soft Comput. › 论文详情
Our official English website, www.x-mol.net, welcomes your feedback! (Note: you will need to create a separate account there.)
Develop and implement unsupervised learning through hybrid FFPA clustering in large-scale datasets
Soft Computing ( IF 3.1 ) Pub Date : 2020-07-03 , DOI: 10.1007/s00500-020-05140-y
Kiran Pandurang Somase , S. Sagar Imambi

Clustering is extensively realistic and considered in computer vision that follows unsupervised learning principles. In this, the performance of a clustering process mainly depends on the feature representation. Generally, the clustering process may have an error rate, and this affects the feature representation. To avoid this, unsupervised Learning (USL) provides an alternative path to obtain the best clusters from the dataset through the optimum features. In this proposed work, the clustering process is done by using a hybrid firefly-based flower pollination algorithm (FFPA). So this clustering process removes the complexity in USL. The better performance is obtained by identifying an essential group from the data to avoid the problems obtained by a USL. In the standard USL, the PCA method is used to minimize a large amount of original data. Here, the features are extracted based on RGB features and Zernike moments, and this is given to the input for the hybrid cluster. Finally, the hybrid convolutional neural network classifier, along with the datasets that are trained from a similar patch manifold, is used to create a label for several datasets. The performance of this proposed method portrays that the local features are effectively clustered from the various datasets by an unsupervised FFPA algorithm. In this work, the unsupervised clustering process with a hybrid classification for the object recognition application is used. In this work, the average accuracy, error rate, and run time are nearly 95%, 73%, and 26 s, respectively.



中文翻译:

通过混合FFPA聚类在大规模数据集中开发和实施无监督学习

群集是非常现实的,并且在遵循无监督学习原则的计算机视觉中被考虑。在此,聚类处理的性能主要取决于特征表示。通常,聚类过程可能有错误率,这会影响特征表示。为了避免这种情况,无监督学习(USL)提供了另一种途径,可以通过最佳功能从数据集中获得最佳聚类。在这项拟议的工作中,通过使用基于混合萤火虫的花授粉算法(FFPA)完成聚类过程。因此,此群集过程消除了USL中的复杂性。通过从数据中确定一个基本组来避免USL所遇到的问题,可以获得更好的性能。在标准USL中,PCA方法用于最大程度地减少大量原始数据。在此,基于RGB特征和Zernike矩提取特征,并将其提供给混合群集的输入。最后,混合卷积神经网络分类器以及从类似补丁流形训练的数据集可用于为多个数据集创建标签。该方法的性能描述了通过无监督的FFPA算法从各种数据集中有效地聚类了局部特征。在这项工作中,使用了针对对象识别应用程序的具有混合分类的无监督聚类过程。在这项工作中,平均准确度,错误率和运行时间分别接近95%,73%和26 s。混合卷积神经网络分类器以及从类似补丁流形训练的数据集可用于为多个数据集创建标签。该方法的性能描述了通过无监督的FFPA算法从各种数据集中有效地聚类了局部特征。在这项工作中,使用了针对对象识别应用程序的具有混合分类的无监督聚类过程。在这项工作中,平均准确度,错误率和运行时间分别接近95%,73%和26 s。混合卷积神经网络分类器以及从类似补丁流形训练的数据集可用于为多个数据集创建标签。该方法的性能描述了通过无监督的FFPA算法从各种数据集中有效地聚类了局部特征。在这项工作中,使用了针对对象识别应用程序的具有混合分类的无监督聚类过程。在这项工作中,平均准确度,错误率和运行时间分别接近95%,73%和26 s。该方法的性能描述了通过无监督的FFPA算法从各种数据集中有效地聚类了局部特征。在这项工作中,使用了针对对象识别应用程序的具有混合分类的无监督聚类过程。在这项工作中,平均准确度,错误率和运行时间分别接近95%,73%和26 s。该方法的性能描述了通过无监督的FFPA算法从各种数据集中有效地聚类了局部特征。在这项工作中,使用了针对对象识别应用程序的具有混合分类的无监督聚类过程。在这项工作中,平均准确度,错误率和运行时间分别接近95%,73%和26 s。

更新日期:2020-07-03
down
wechat
bug