The presence of abnormal ST segments in electrocardiograms (ECGs) is associated with a variety of prevalent medical conditions. However, current technologies for automatic arrhythmia detection predominantly focus on QRS complexes, and there is a notable lack of research dedicated to identifying anomalous ST segments. To effectively and autonomously identify unusual cardiac activity based on ECGs, we propose a novel methodology for ST anomaly detection that combines a modified sequential transformer with convolutional techniques. This approach integrates an inverted transformer and channel prior attention into a unified network, resulting in high detection performance with a reduced error rate while utilizing single-lead diagnostics. On our collected dataset, comprising over 180,000 authentic clinical records, the overall detection accuracy reached an impressive 96.14%, with an outstanding F1 score of 96.17%. For the PTB-XL database, we achieved an accuracy of 88.58% and an F1 score of 85.02%. These remarkable findings surpass the accuracy and generalizability of existing methodologies, thereby equipping clinicians with an advanced tool for detecting abnormalities in ST segments.