In this research, a novel deep-learning approach is introduced for detecting and analysing defects in golf swing actions, aiming to enhance performance and prevent injuries in sports biomechanics. Utilizing Google MediaPipe for real-time anatomical landmark estimation, a dataset of 160 golf swings was created under controlled conditions, capturing a range of swings including common errors. Joint angles were calculated from anatomical landmarks to provide comprehensive feature sets. To efficiently manage the high-dimensional data, autoencoder models were employed to compress the features while preserving critical information. The primary innovation is the development of the SCA-LSTM model, which integrates Long Short-Term Memory (LSTM) networks with Squeeze-and-Excitation (SENet) and Contextual Transformer (CoT) attention mechanisms. The integration of SENet and CoT attention mechanisms with LSTM networks significantly enhances feature representation and contextual understanding, resulting in superior performance in analysing complex motion sequences. Experimental results demonstrate that the proposed model achieves a classification accuracy of 96.88%, substantially higher than the 87.5% accuracy of baseline LSTM models. The findings underscore the model's effectiveness in detecting and analysing golf swing defects, suggesting broad applicability to various sports and training scenarios within Human Motion Modelling and Analysis (HMMA).