Label data cleaning device and method

The invention discloses an annotation data cleaning device which comprises an annotation database, an algorithm model structure, a data equally-dividing module, a sub-data-set setting module, a training module and a testing module. The sub-data set setting module sequentially selects one sub-data set from the N sub-data sets as a sub-training set, the remaining sub-data sets are sub-test sets respectively, and one sub-training set and the N-1 sub-test sets form a training test group. And the training module adopts the sub-training sets in each training test group to train an algorithm model structure. And the test module adopts each sub test set to test the corresponding trained algorithm model in sequence and form a test result, collects all error examples from the test result and cleansthe labeled data corresponding to the error examples. The invention further discloses an annotation data cleaning method. According to the invention, automatic cleaning of labeled data can be realized, and the cleaning efficiency and the cleaning quality can be improved.