Image annotation method based on semi-supervised learning

The invention discloses an image annotation method based on semi-supervised learning, and the method comprises the steps: designing different classifiers for different types of samples, training the classifiers through a part of annotated samples, carrying out the voting of the results of the different classifiers, selecting a type with the highest accuracy, and carrying out the annotation of an unknown sample. The influence caused by wrong classification is reduced; the samples in each category obtained by the classifier and the labeled samples in the corresponding category are subjected to random linear mixing operation, so that the result of error classification also contains the features of the corresponding category, and a new thought is provided for semi-supervised learning in the fields of deep learning and machine learning. 本发明公开了基于半监督学习进行图像标注方法，针对不同类别的样本设计不同的分类器，利用已经标注好的部分样本来训练分类器，并且对不同分类器的结果进行投票，选择出准确率最高的类别，从而对未知样本进行标注。并且为了降低错误分类带来的影响，将分类器得到的每一个类别中的样本与标注的相应类别中的样本进行随机线性混合操作，使得错误分类的结果中也含有对应类别的特征，为半监督学