Pediatric Sleep Apnea-Hypopnea (SAH) significantly impacts children's health. The standard diagnostic method, Polysomnography (PSG) test, is effective but uncomfortable, particularly for children. This research introduces a less intrusive approach to assess the severity of pediatric SAH using blood oxygen saturation (SpO2) signals. Leveraging the Inception V3 model, we transform onedimensional (1D) SpO2 signals into two-dimensional (2D) spectrograms for a detailed analysis through transfer learning. Applied to the CHAT dataset, our method achieved notable diagnostic accuracy, demonstrating proficiency in estimating the Apnea-Hypopnea Index (AHI). The model exhibited 74.11% accuracy across four SAH severity classes and a kappa score of 0.6 on the test set. With accuracies of 85.8%, 90.5%, and 96.4% for common AHI thresholds, it proved effective in identifying pediatric SAH, contributing significantly to the simplification and improvement of diagnosis with a non-invasive, childfriendly approach. Despite challenges in AHI estimation, especially for severe cases, this study advances efforts towards enhanced pediatric SAH diagnosis.