Abstract: The traditional manual inspection mode has problems such as poor timeliness, strong subjectivity of data, and high cost of large-scale monitoring in building surface damage monitoring. To address these technological bottlenecks, this paper proposes an intelligent recognition method for building surface damage based on the fusion of multi temporal images from unmanned aerial vehicles and long short-term memory networks (LSTM). This study used a multi rotor drone equipped with a hyperspectral camera to obtain multi-phase image data of a target building complex (with a spatial resolution of 20 cm~80 cm), and constructed a heterogeneous dataset containing geometric deformation features and texture changes. In response to the complexity of building structures and the spatiotemporal heterogeneity of damage modes, temporal modeling is used to achieve dynamic correlation analysis of multi temporal images, effectively capturing the evolution process of damaged areas. The research results indicate that when the spatial resolution of the training set is optimized to 60 cm, the model achieves the optimal performance indicators. At this scale, it is possible to preserve the detailed features of the edges of building components while effectively suppressing the interference of image noise, thereby achieving pixel level localization and sub meter level accuracy measurement of damaged areas. The intelligent recognition method based on the fusion of unmanned aerial vehicle (UAV) multi temporal images and LSTM proposed in this article can effectively solve the shortcomings of traditional manual inspection mode and provide an efficient and accurate technical means for monitoring building surface damage, which has important application value.