This research developed a model for scoring athlete technical actions using deep learning. It combines CNNs and LSTMs to extract spatial and temporal features in sports actions and accurately score tasks. The model utilizes standardized sports key point data, generates spatial features via multi-layer convolution, employs LSTM to capture action’s temporal dynamics, and eventually produces scoring outcomes. Through experimental verification on multiple public data sets, the model performs well in multiple evaluation indicators such as mean square error (MSE), mean absolute error (MAE) and determination coefficient ( \(R^{2}\) ), proving its effectiveness and wide applicability in different sports action scoring tasks. This research also examines the significance of data preprocessing, model architecture design, and hyperparameter tuning in enhancing model performance. It also confirms their potential usefulness in real-time scoring for sports events and training. The findings of this research establish a strong base for the future utilization of deep learning technology in sports, and offer a valuable guidance for additional studies in associated areas.