Respiratory diseases in neonates are among the leading causes of neonatal illness and death, particularly in developing countries. Prompt diagnosis and treatment of these conditions are essential. Thermal imaging emerges as a non-invasive and radiation-free diagnostic approach, utilizing temperature variations and thermal symmetry monitoring as tools in medical diagnostics. This study explores the detection of neonatal respiratory abnormalities using artificial intelligence applied to limited thermal imaging data. Convolutional Neural Network (CNN) models, while highly effective for classification tasks, typically require extensive and balanced datasets. However, obtaining sufficient neonatal thermal imaging data can be challenging due to the delicate nature of care in neonatal intensive care units. To address this limitation, the study incorporates a robust deep learning framework alongside various data augmentation techniques to enhance classification outcomes. Neonates with respiratory abnormalities were grouped into one category, while those with cardiovascular and abdominal issues were grouped into another. Results showed that data augmentation, which increased the dataset size by four times, improved classification accuracy from 84.5% to 90.9%. With deterministic feature extraction as well as the incorporation of supervised learning, this paper presents a hybrid method for data processing acquired through UAVs.