<p>Chronic Obstructive Pulmonary Disease (COPD) is a prevalent respiratory condition characterized by a gradual decline in airflow and persistent breathing difficulties. Timely and precise diagnosis is essential for effective patient care and for minimizing healthcare expenditures. Although spirometry is commonly employed for COPD diagnosis, it is relatively invasive and demands specialized clinical expertise. Lung sound analysis provides a non-invasive, affordable, and dependable approach for detecting COPD. This research proposes a ViT-based hybrid feature framework for COPD classification using lung sound recordings from the ICBHI 2017 dataset. The proposed framework is organized into four key phases: preprocessing, feature extraction, model development, and performance evaluation. Preprocessing steps include resampling, filtering, amplitude normalization, and segmentation, along with data augmentation to enhance robustness. A hybrid stacked representation was generated by fusing diverse time–frequency features, including MFCC, chroma, and spectral contrast, to enhance the richness of the extracted information. Vision Transformer (ViT) architecture was employed for the classification task, which incorporated feed-forward layers, residual connections combined with layer normalization, and multi-head self-attention techniques. This architecture efficiently models both local and global relationships within the stacked features, thereby enhancing the quality of representation learning. Experimental findings validated the efficiency of the developed framework, achieving performance rates of 97.85%, 98.12%, 97.98% and 98.00% for precision, recall, F1-score, and accuracy, respectively. The study demonstrates that combining hybrid stacked features with ViT provides a powerful solution for COPD detection, supporting the development of automated, non-invasive diagnostic systems in respiratory healthcare.</p>