This paper introduces a novel image-level ensemble learning method called Random Color Erasing (RCE) to enhance the robustness of deep learning models in scenarios with color variations. RCE selectively erases partial or complete color information in training data without disrupting the original image structure, achieving a balanced weighting of color features and other features within the neural network. This approach mitigates overfitting and enhances the model's ability to handle color variations, improving overall robustness. The method is characterized by robust interpretability and is evaluated on various tasks such as person re-identification and semantic segmentation. Experiments across different datasets and baseline models demonstrate the effectiveness of RCE, showing significant improvements over existing methods, particularly in cross-domain testing. The paper also provides an analysis of the relationship between RCE and the generalization ability of neural networks, highlighting the intrinsic reasons for the superior performance of networks trained with RCE.This paper introduces a novel image-level ensemble learning method called Random Color Erasing (RCE) to enhance the robustness of deep learning models in scenarios with color variations. RCE selectively erases partial or complete color information in training data without disrupting the original image structure, achieving a balanced weighting of color features and other features within the neural network. This approach mitigates overfitting and enhances the model's ability to handle color variations, improving overall robustness. The method is characterized by robust interpretability and is evaluated on various tasks such as person re-identification and semantic segmentation. Experiments across different datasets and baseline models demonstrate the effectiveness of RCE, showing significant improvements over existing methods, particularly in cross-domain testing. The paper also provides an analysis of the relationship between RCE and the generalization ability of neural networks, highlighting the intrinsic reasons for the superior performance of networks trained with RCE.