Traditional single-wavelength or thermal imaging methods for aircraft wing icing detection rely on complex data processing， making it difficult to achieve rapid and stable inversion of wing ice thickness. Additionally， measurement errors arise from variations in the angle of incidence of the detection light caused by the curved surface structure of the wing. To address these issues， this paper proposes a non-contact wing icing detection method based on an infrared dual-wavelength system. This method establishes an ice thickness measurement model based on the Lambert-Beer law and the structural characteristics of the wing surface. It introduces an adaptive angle correction factor， ， to correct measurement errors caused by the curvature of the wing. Furthermore， by optimizing the calibration process， the method significantly improves detection efficiency， enabling rapid and accurate detection of wing ice thickness. Simulation and experimental validation demonstrate that the coefficient of determination for ice thickness measurements from this mathematical model exceeds 0.9 across all four detection angles， with a maximum root mean square error RMSE of 0.4037. Within the detection angle range of 15° to 18°， the relative error ranges from 5.2% to 13.1%， with the lowest error （5.2%） occurring at 15° with the highest error occurring at 18° （13.1%）. This method establishes a physical model that integrates geometric optical paths with dual-wavelength absorption. Through adaptive angle correction， it can meet the requirements of most application scenarios for wing icing detection. The method achieves high inversion accuracy in wing icing detection， fundamentally overcoming measurement errors caused by the curvature of the wing surface， and provides a new technical approach for the development of high-precision icing detection systems.