Automating Serve Shot Analysis in Table Tennis with CNN-LSTM Technology

In this research, a CNN-LSTM architecture is developed to identify serve shots of table tennis. The system uses convolutional neural networks and long short-term memory to provide spatial and temporal features. The model significantly improves the serve categorization, resulting in 93%, as opposed to the machine learning classification, independent CNN (88%), and LSTM techniques (86%). The serves in the dataset cover backspin, topspin, and sidespin serve, which is labelled with a flag of serve type and a flag of its effectiveness. The CNN module acquires spatial features from the frames of the videos, which means gauging the players' movements and the orientation of the… At the same time, the LSTM network determines the temporal patterns of sequences of serves. The confusion matrix indicates that the categorization of backspin, topspin and sidespin serves was done with an accuracy of 85%, 92% and 88%, respectively. The temporal analysis presents a high positive correlation (of 0. 76) between the velocity of the serve and its effectiveness; the findings are helpful for the players and coaches. The training and validation graphs for accuracy and loss prove that the model is well trained. By epoch 20, the training accuracy is 97 per cent, accompanied by validation accuracy of 96 per cent, thus little overfitting. The ROC curve, in conjunction with the AUC of 0, indicates that 95 confirms the fact that the model demonstrated outstanding capacity to differentiate between successful serves and less successful ones. Thus, the present research provides a novel intervention to offer specific descriptions of shots in table tennis, providing accurate, versatile., and functional data.