Forehand to Deep Learning: A CNN-LSTM Framework for Serve Shot Analytics in Table Tennis

It introduces a novel method for detecting and recognizing the table tennis serve-shot motions based on the advanced convolutional and Long Short-Term Memory models. The model is designed to classify serve shots into five categories: side-spin, backspin, shots with no spin, slicing, and illegal service. It consists of CNNs for precision image classification and the LSTM network for considering the time movement of people through a sequence of shots. Experimental data proved the viability of a model, so the overall accuracy was achieved at ${9 0. 3 9 \%}$. The accuracy metrics for the varied classes can stretch from 84.6% to ${9 4. 2 2 \%.}$ Another key indicator is the high recall rates for Serve shots. This table has an astonishing rate of ${9 7. 0 3 \%}$ for class 3, demonstrating how the model can effectively recognize serve shots. The high F1 scores illustrate the balance between the Precision and the Recall, the values of which exceed 87% for all instances. The model’s performance is guaranteed as performance evaluations are based mainly on macro, weighted, and micro averages. We usually have 90-95 percent accuracy, recall, and F1-Score scores. The experiment proves that the classifier finds the answer to the issue, which is indeed a significant achievement considering the job’s toughness and numerous data. The CNN-LSTM model’s fantastic accuracy and dependability are a reference tool for coaches and athletes they can rely upon. While some argue that it can help trainers and give a macro view into the course, others cannot stop them for fear that this will create doubts about the authenticity of the competition.