针对酸蚀后铸轧铝合金表面晶粒细小且取向复杂，导致难以仅依赖光学这一不稳定特征准确提取晶粒取向信息的问题，本研究提出了一种综合组织光学与高度特征来获取铸轧铝坯晶粒取向特征的方法。该方法首先在获得晶粒取向与酸蚀组织表面光学与高度原始测量数据的基础上，利用特征工程，构建了包含灰度共生矩阵与几何曲率在内的多维特征集以获取更多的酸蚀表面微观形貌信息，并且通过后处理对构建的堆叠集成模型预测输出结果进行训练优化。选用的堆叠模型在测试集中的正确率为78.3%，宏平均 F1-score为0.784。在验证实验中，预测结果在空间分布上与真实取向具有较好一致性，正确率为80.1%，并顺利实现更大区域晶粒取向特征预测。最后，采用SHAP可解释性模型（SHapley Additive exPlanation）佐证模型的科学合理性。SHAP分析结果表明，铸轧铝酸蚀微观形貌中的光学特征以及纹理特征在模型中能够有效反映晶粒取向差异，高度特征在模型中也主要起着调节、稳定模型的作用，三者结合共同作用实现了模型良好的预测性能。本研究为利用铸轧铝合金及类似合金酸蚀表面组织形貌获取晶粒取向信息以及进一步高效实现大区域晶体学取向定量识别提供了可行技术路径。

The defect of coarse grains on the surface of cast-rolled aluminum slabs significantly impacts the yield of subsequent rolling processes. These coarse grains manifest not only in size variations but also in pronounced orientation differences. Therefore, investigating the orientation characteristics of surface grains in cast-rolled slabs is crucial for improving the microstructure and surface quality of cast-rolled products. Traditional inspection methods, which primarily rely on macroscopic observation after acid etching, struggle to quantitatively reflect grain orientation variations. To address the challenge of accurately extracting grain orientation information solely through optical characteristics—which are inherently unstable—due to the fine grain size and complex orientation of cast-rolled aluminum alloys after acid etching, this study proposes a method that integrates optical and height features to obtain grain orientation characteristics of cast-rolled aluminum slabs. This method first utilizes feature engineering to construct a multidimensional feature set—including grayscale co-occurrence matrices(GLCM) and geometric curvature—based on raw optical and height measurements of grain orientation and microstructure surfaces. This approach extracts additional etching micro-morphology information. Subsequently, post-processing trains and optimizes the prediction outputs of the constructed stacked ensemble model. The selected stacked model achieved an accuracy of 78.3% on the test set, with a macro-average F1-score of 0.784. In the validation experiments, the predicted results showed good consistency with the actual orientations in spatial distribution, achieving an accuracy rate of 80.1% , and successfully enabling the prediction of grain orientation features across larger regions. Finally, the SHAP (SHapley Additive exPlanation) interpretability model was employed to substantiate the scientific validity of the model. SHAP analysis results indicate that optical features and texture features within the microstructure of acid-etched cast-rolled aluminum effectively reflect grain orientation variations in the model. Height features primarily serve to regulate and stabilize the model. The combined action of these three elements collectively achieves the model's excellent predictive performance. This study provides a viable technical approach for obtaining grain orientation information from the surface microstructure of acid-etched cast-rolled aluminum alloys and similar alloys, and for efficiently achieving quantitative identification of crystallographic orientation across large areas.