Sports Action Recognition and Standardized Training using Puma Optimization Based Residual Network

Trainers/coaches in sports show high standards for their skill because they provide key training tips to students for achieving own goals by focusing their performances. To automate that activity, applying artificial intelligence technology to the field of sports training, a sports action recognition combined training method is proposed utilizing an Optimized Deep Learning (DL) mechanism named Puma Optimization based Residual Network (PO-ResNet) model. The ResNet is used for recognizing the action and the PO is used for tuning the hyperparameters of ResNet to improve recognition accuracy. Utilizing wearable sensors, the system records the user’s motion data, then the mix-max normalization technique is used to preprocess the signal by removing unwanted noises to perform Sport action recognition effectively. Then, the proposed PO-ResNet optimization model extracts the features and recognition the user’s activities effectively. Subsequently, the PO-ResNet enables the user to standardize sports combination training. The suggested PO-ResNet technique produced good recognition accuracy of 98.97% and outperforms conventional approaches such as Convolutional Neural Network, Bidirectional Long Short Term Memory and ResNet (BiLSTM+ResNet) and Gated Recurrent Unit (GRU).