Abstract: The rapid development of digital image technology has led to its widespread application in the field of automatic crack recognition.However, the accuracy of crack recognition using a single algorithm model in complex backgrounds still needs improvement.To address this, we utilized digital orthophoto maps (DOMs) obtained through UAV oblique photography as the data source for automatic crack identification research.Firstly, considering the imaging characteristics of different crack types, three algorithm models (threshold segmentation, edge detection, and supervised classification) were employed for preliminary crack identification, using image pixel gradient, grayscale value, and RGB value as indicators.Next, the initial extraction results were processed by morphological restoration and two filtering algorithms of direction and length to handle background noise.Finally, a minimum-risk Bayesian probability model fusion method was proposed to integrate the recognition results from different models at a decision level.The engineering application results demonstrated that: the three algorithm models effectively extracted cracks on reservoir bank slopes, each achieving an accuracy rate of over 70%.Model fusion minimized the omission of true cracks, with the final fused model achieving a recognition accuracy of 95.4%.The fusion algorithm successfully distinguished target cracks from background information in reservoir slope crack identification, significantly reducing image noise and improving recognition precision.