<p>Obstructive sleep apnea–hypopnea syndrome (OSAHS) has a significant impact on public health and safety. Polysomnography, which is currently used as the gold standard for its detection, has drawbacks such as being time-consuming, having issues with data accuracy, and requiring specialized personnel for interpretation. Current methods are mainly divided into two categories: traditional machine learning methods that manually extract features and then input them into classic models for classification, and deep learning models that automatically extract features for training. However, due to the diversity of signals and different feature extraction methods, the classification task of OSAHS still poses significant challenges. To address these difficulties, this paper selects thoracoabdominal movement and oronasal airflow signals from polysomnography results as inputs and proposes a dual-channel spatiotemporal attention mechanism model based on the Transformer architecture for feature extraction. It then combines this with a gating mechanism to selectively integrate features and ultimately obtain classification results. We used the classic dataset SHHS and a dataset provided by Shanghai Sixth People’s Hospital for retrospective testing. The experimental results show that our proposed model successfully achieved the expected goals, with an accuracy of 0.9, a sensitivity of 0.9, a specificity of 0.903, and an AUC curve value of 0.903.</p>