<p>The advancements of artificial intelligence technology, particularly the deep learning model’s emergence, have improved the accuracy of monitoring, evaluation, and prediction of agricultural meteorological disaster. Deep learning models have the advantages of strong learning ability, great portability, wide coverage, and strong adaptability. Accurately evaluating the impacts of freezing damage on winter wheat using deep learning model is significant for further improving the defense capability of agricultural meteorological disasters and realizing the sustainable development of agriculture. The dynamics in spring freezing damages and their impacts on winter wheat under various levels of freezing damage during 1981-2020 in the Huang-Huai-Hai (HHH) region of China were evaluated based on the Long Short-Term Memory (LSTM) deep learning model and ground observations. Our research results indicated that the frequencies of spring freezing damage of winter wheat from 1981 to 2020 at different freezing damage levels in the HHH region all showed decreasing trends with significant fluctuations. Over the past forty years, the winter wheat in the HHH region was primarily threatened by moderate freezing damage, followed by severe freezing damage, and finally mild freezing damage. Furthermore, some clear spatial differences in spring freezing damages of winter wheat during different developmental stages were found. Particularly, the winter wheat planting areas of North of 34°N and East of 114°E were more prone to severe freezing damages. In addition, the impacts of spring freezing damage on winter wheat varied greatly at different development stages. The winter wheat was more severely affected during the jointing stage and the heading-flowering stage, followed by the regreening stage, and finally the booting stage. We expect these findings will improve our understanding on the impacts of freezing damage on winter wheat under climate change.</p>