Damage identification is one of the most important aspects of structural health monitoring. In order to further improve the efficiency and accuracy of damage identification, a two-stage structural damage identification method based on the improved modal strain energy index and DBO-BP (Dung Beetle Optimization-Back Propagation) neural network is proposed. Firstly, the improved normalized damage index of modal strain energy is used for the damage localization analysis of the structure. Then, taking the average change rate of the structure unit modal strain energy as the input parameter, the stiffness reduction coefficient of the damaged unit as the output parameter, and utilizing the DBO algorithm which is an improved version of the optimal latin hypercubic method, the weights and thresholds of the BP neural network are optimized to perform the structural damage quantitative analysis. The concrete slab structure and flat rigid frame structure are used as exemplary models for damage identification verification. The results show that the proposed method is accurate in damage location identification, with high calculation efficiency for the degree of damage and small identification error as low as 0.4 %, exhibiting excellent damage identification performance.