To address the issue that the reliance on a single data source in current sea ice thickness detection limits the further improvement of sea ice thickness inversion accuracy.This paper proposes a sea ice thickness inversion method based on segmented feature fusion of heterogeneous data，the experiment uses Sentinel-1 synthetic aperture radar （SAR） data and ERA5 reanalysis data，by dividing sea ice thickness intervals （e.g.， （0， 1.5） m， （0， 2） m， （0， 3） m）， optimal feature combinations are selected for sea ice in different segmented intervals，a sea ice thickness inversion model based on stacked ensemble learning is constructed， which realizes complementary advantages through the series-parallel cascade of multiple base models and meta-models，this approach fully explores the hidden correlations between heterogeneous features and sea ice thickness to achieve accurate inversion of segmented sea ice thickness.The results show that compared with other traditional machine learning methods， this method achieves better overall inversion performance across different segmented intervals，notably， the interval of （0， 1.5） m exhibits the best performance， with a coefficient of determination （R²） reaching 0.923 and a root mean square error （RMSE） as low as 0.089 m.The study demonstrates that segmented feature optimization and heterogeneous data fusion can effectively improve the inversion accuracy of sea ice thickness， verifying the advantages of the proposed stacked ensemble learning model in heterogeneous data fusion. This research provides a new method for achieving high-precision inversion of sea ice thickness.