基于集成GMM聚类的少标记样本图像分类

Classification of Few Labeled Images Based on Integrated GMM Clustering

为了提高卷积神经网络训练的分类器分类准确率,往往需要大量的已标记数据,但有时已标记数据并不容易获得。针对少标记样本图像分类问题,提出基于集成GMM聚类与标签传递思想的解决方案,通过一定的规则给未标记数据赋予标签,将未标记数据转换成已标记数据用于模型的训练。在手写数字识别数据集上进行实验,结果表明新算法在少标记样本的情况下,结合集成 GMM 聚类的方法比只采用有标记样本训练得到的模型分类准确率有着较大提高,验证了该算法的有效性。

In order to improve the classifier classification accuracy of by using convolutional neural network training, a large amount of labeled data is often required, but sometimes labeled data is not easily obtained.This paper proposes a solution based on the idea of integrated GMM clustering and label delivery for classifying images with few labeled samples, assigning tags to unlabeled data through certain rules, and converting unlabeled data into labeled data for training of the model.In this paper, experiments are performed on hand-written digital recognition data sets. The results show that the present algorithm has a great improvement in the accuracy of model classification comparing with the method of using only labeled samples in the case of few labeled samples. The effectiveness of the present algorithm is validated.