Aiming at the problem that the unbalanced characteristics of the data returned by the human body posture monitoring sensor affect the classification performance

a human body attitude classification algorithm based on unbalanced data classification is proposed. Firstly

according to the characteristics of the data returned by the attitude monitoring sensor

a noise sample identification algorithm based on the idea of K-means is proposed.Secondly

aiming at the imbalance of the sample set

operate on a few kinds of sample sets by introducing the classical oversampling algorithm SMOTE. Finally

using the advantages of the Adaboost learning framework

the balanced sample set is trained to obtain the final classification model. Choose G-mean

F-value

and AUC as the evaluation indicators of the classification model

and compares it with the three classic unbalanced classification models

CUS-Boost

SMOTEBoost

and RUS-Boost on the ARe Mr human pose dataset. The validity and accuracy of the proposed human pose classification algorithm based on unbalanced data classification are verified.