Under the coupled effects of internal fracture structures and complex loading conditions, the roofs of metal mine goafs are prone to stress redistribution and intensified local damage, which reduce their overall load-bearing capacity and increase the risk of instability. To investigate the spatial evolution characteristics of internal damage and crack networks in goaf roofs under stress, a series of overload failure tests were conducted on physical similarity models of goaf roofs designed based on moderately thick plate theory. Acoustic emission (AE) monitoring was employed to capture the spatiotemporal evolution of crack propagation during roof failure, revealing the evolution patterns of AE energy, event density, and crack types at different damage stages. On this basis, an adaptive kernel density estimation (AKDE) method was introduced to achieve three-dimensional visualization and identification of roof damage zones. In addition, a continuous crack network model based on discrete fracture network (DFN) theory was constructed using AE events. The crack network branching coefficient exhibits a characteristic evolution trend of gradual increase, peak occurrence, and subsequent decline, quantitatively describing crack propagation paths and the evolution of network topology. These features effectively reflect the progressive transition of the goaf roof from localized damage to global instability. The proposed approach provides new quantitative methods and theoretical support for internal damage identification, failure warning, and safety assessment of goaf roofs.