A global health concern, chronic obstructive pulmonary disease (COPD) demands early detection and intervention for effective treatment. Leveraging the extensive 12-channel lung sound dataset "RespiratoryDatabase@TR", our study establishes a robust multiclass COPD severity diagnostic system. We employ a rigorous feature extraction procedure, including spectrogram, Mel spectrogram, and chromogram analysis, alongside specific data preprocessing and augmentation methods. The RESNET50 model is chosen for training, ensuring precise classification across COPD severity levels. Our findings underscore the significance of sound-based prognoses, particularly in early COPD diagnosis. With an estimated 251 million people globally affected by COPD, innovative solutions like ours are crucial. The amalgamation of extensive datasets and advanced machine learning techniques holds the promise of transforming COPD diagnosis and treatment on a global scale, improving the lives of millions affected by this illness.