The integrity of the corrosion protection layer in natural gas pipelines is crucial for pipeline safety, as cracks are a major hidden hazard that can lead to leaks and accelerate corrosion. This paper employs an air-coupled ultrasonic intra-resonance detection technique, processing the ultrasonic detection signals using ensemble empirical mode decomposition (EEMD) to separate the intrinsic mode function (IMF) components. These components are then analyzed for their correlation with the original signal and subsequently processed using wavelet threshold denoising technique. The denoised signal is reorganized to improve clarity, and six dimensional feature quantities are extracted as inputs to the extreme gradient boosting (XGBoost) model. To further optimize the model′s performance, key parameters such as the number of iterations, tree depth and learning rate of XGBoost are optimized using the Newton-Raphson (NRBO) algorithm to achieve the best recognition efficiency and accuracy. The results show that the method achieves a damage recognition accuracy of 95.83% for the corrosion protection layer, with an average relative error of only 6.3% in crack length prediction. This demonstrates excellent anti-interference ability and high accuracy. The method provides a new idea for natural gas pipeline anti-corrosion layer detection, contributing to improve pipeline safety, reduce maintenance costs and extend pipeline service life.