Enhancing behavior classification of children in dynamic interaction scenes through improved DCNN model.

The rapid development of society makes people pay more attention to the quality of the environment for children's growth. However, due to the differences of young children, different environments are often needed for cultivation in dynamic interaction scenarios. Therefore, the authors propose an environment creation method for children's behavior classification to improve the quality of children's growth environment. Taking the video data of children for a period of time as input, the encoder and decoder are designed to classify children's behavior and obtain behavior characteristics. After the input image is processed by the backbone network DCNN, two outputs are obtained, which are four times of shallow features and 16 times of high-level features. Aiming at the semantic gap between environmental features and children's behavior features, the DenseNet model is used to remove the semantic difference between children's behavior features and environmental features, and the similarity between the two features is fitted as much as possible. The dense blocks obtained by different expansion factors of the network are used for feature connection, so that the model is suitable for feature similarity calculation of different modes. The experimental results show that this method can accurately classify children's behavior, and the F value is more than 70%, which can provide prerequisites for children's environment creation. This environment creation model can clearly point out the suitable environment for children and provide a guarantee for children's growth.