Abstract: The strip crown plays a crucial role in determining the quality of products in strip hot rolling. Hence, achieving precisely hot-rolled strip crown diagnosis is important to improve the control capability of hot rolling. Since the hot rolling process features nonlinearity, heredity, and strong coupling, the diagnosis of strip crown is an imbalanced problem with complex decision boundaries. Existing crown prediction models tend to learn more information from the majority class, but ignore the data of the strip with unqualified crown. To overcome this limitation, a hot-rolled strip crown diagnosis model based on the fusion of hybrid resampling and cost-sensitive is proposed, and the cost-sensitive factor is obtained by artificial hummingbird algorithm. Some advanced machine learning models are selected as comparison models. The experimental results demonstrate that the proposed model outperforms all comparison models with the A_UC of 0.889 and defect recall of 0.870. Moreover, the testing time of the proposed model is only 0.0059 s. After the madel was applied to online diagnosis,the detection rate of defect crown strips increased from 82% to 88%,and the crown compliance rate of strips improved from 59% to 71%.