THE IMPACT OF EXECUTIVE FUNCTION AND AEROBIC EXERCISE RECOGNITION IN OBESE CHILDREN UNDER DEEP LEARNING

It was to investigate the impact of an optimized OpenPose motion recognition model assisted by spatial transformer network (STN) and Lucas–Kanade optical flow algorithm on the functional capacity of children with simple obesity during aerobic exercise. Initially, a motion recognition model based on STN and Lucas–Kanade optical flow algorithm optimization was constructed. The COCO-WholeBody Dataset was utilized as the training set for the model. The performance of the model before and after optimizations was evaluated to obtain the optimal parameters. Simultaneously, 200 cases of children with simple obesity were recruited as the research subjects and randomly rolled into a control group (CG, conventional aerobic exercise) and an experimental group (EG, aerobic exercise assisted by the optimized OpenPose motion recognition model based on STN and Lucas–Kanade optical flow algorithm). The body composition and functional capacity of the children in both groups were compared. Following the optimizations, the OpenPose motion recognition model exhibited higher accuracy in keypoint detection and greater similarity in pose estimation compared to its pre-optimized state. Moreover, keypoint localization error decreased post-optimization. Lower depths of the convolutional neural network (CNN), larger input image sizes, and batch sizes were associated with improved performance across evaluation metrics. The optimal parameters for the model were determined as a CNN depth of 18 layers, input image size of [Formula: see text], and batch size of 128. The optimized OpenPose model significantly enhanced processing time, memory consumption, frame rate, and accuracy through the integration of STN and the Lucas–Kanade optical flow algorithm, demonstrating particularly outstanding performance in real-time motion recognition. Although the performance slightly decreased on the dataset of children with obesity, the model exhibited good adaptability and high accuracy across multiple datasets. In comparison to the CG, children in the EG showed marked reductions in body weight, fat content, BMI, and body fat percentage ([Formula: see text]). However, there were neglectable differences observed in muscle mass, inorganic salts, and protein content ([Formula: see text]) between the two groups. Furthermore, children in the EG exhibited higher rates of inhibition, refreshment, and transformation functions, along with lower reaction times, compared to the CG ([Formula: see text]). The OpenPose motion recognition model optimized with STN and Lucas–Kanade optical flow algorithm not only exhibits superior performance in keypoint detection and pose estimation but also positively influences the physical health and functional capacity of children.