RANSAC-Enhanced Calibration of Alignment Errors in Acousto-Optic Tunable Filter Imaging Systems

Accurate calibration of alignment errors is critical for model-based spectral correction in acousto-optic tunable filter (AOTF) imaging systems, especially for wide field-of-view applications. In practice, experimental frequency response data are often contaminated by outliers originating from laser instability, field-dependent diffraction efficiency variations, and non-ideal peak fitting, which severely bias conventional least-squares estimation due to the strong nonlinearity between alignment errors and image-plane frequency responses. In this work, we demonstrate that coupled alignment errors give rise to a structured single-minimum frequency response surface governed by system geometry, and exploit this physical constraint to develop a model-guided robust calibration framework. A Random sample consensus (RANSAC)-based estimation strategy is employed to identify and exclude outlier measurements without assuming a specific noise distribution, followed by constrained least-squares optimization on the purified dataset. Experimental validation on a collimated mid-wave infrared AOTF imaging system shows that the full-field frequency response modeling error is reduced from 2.4% to below 0.3% after calibration. The proposed method significantly enhances the robustness and accuracy of alignment error compensation, providing a practical solution for high-precision AOTF spectroscopic imaging under real-world measurement uncertainties.