Modeling and Predictive Analysis of the Validity of Children's Behavior Observation in Preschool Education based on Action Recognition Network Algorithm

In view of the high complexity of RGB image background and the lack of depth information, this paper proposes the M-CNN model, which fuses the RGB image information of static human actions and the skeleton point information, and uses the skeleton point information to effectively make up for the RGB image information in the image. For the lack of in-depth information, formulate a framework for the development of children's behavior observation ability for preschool education students; add a professional skills course of "children's behavior observation and evaluation"; add "children's behavior observation ability" task assessment in educational probation; strengthen the penetration of other related courses, based on the action recognition network to establish a prediction model of children's behavior observation, the efficiency is increased by 7.3%.