Abstract: The problem of root rock loss is the key to the stability study of groin bank root-rock pile. By timely identifying the precursors of a large number of root rock loss, danger situation can be avoided. The prediction of root stone loss cannot be based on subjective assumptions, and the key to this study is to informationization the judgment basis of traditional manual detection methods. On the basis of real-time monitoring of 171 groin bank root-rock piles using image recognition technology, the limited deformation characteristics of the root rock piles are processed according to the principles and methods of grey theory. Then, by iteratively comparing with the graphics of the defect morphology database, more information is generated to make the precursor problem of large-scale loss of the root-rock as clear as possible. The research results indicate that the system model constructed using grey theory can objectively reflect the correlation between the overall stability of the lost root-rock and the root-rock pile, as well as the development trend of the deformation of the rootrock pile. This helps flood departments to quickly and accurately predict the stability of the groin bank root-rock pile.