Breathing is considered a crucial physiological metric when monitoring human vital signs. In resource-constrained environments with limited access to trained medical professionals, the automated analysis of abnormal breathing patterns can offer significant advantages to healthcare systems. In this research paper, we have implemented hybrid deep learning models to classify individuals' breathing patterns using two types of features: signal image-based features and spectrogram image-based features. We used the Sleep Breathing Detection Algorithm (SBDA) to preprocess the data to extract the actual breathing signals from ultra-wideband (UWB) radar. Subsequently, the signals were transformed into signal images and spectrogram images to serve as input features for the hybrid deep learning models. Additionally, two other deep learning models were employed to validate the performance of the proposed approach. To evaluate the effectiveness of our method, we employed five performance metrics, including accuracy, precision, recall, specificity, and F1-score. The overall results clearly demonstrated that our proposed method outperforms the two alternative deep learning models that were utilised for normal and abnormal breathing classification. These findings highlight the superior performance of our hybrid deep learning approach in accurately distinguishing between normal and abnormal breathing patterns. By automating the analysis of breathing patterns, our method shows great potential for enhancing healthcare systems, particularly in settings where resources and trained medical professionals are limited.