Prior-Guided SEM Image Segmentation Enables Quantitative Morphology Assessment and Process Optimization in Perovskite Solar Cells

Perovskite film morphology strongly affects charge transport, nonradiative recombination, and power conversion efficiency, yet scanning electron microscopy (SEM) remains largely qualitative. Here, a prior-guided SEM analysis framework converts routine micrographs into class-resolved masks, quantitative morphology descriptors, and a compact SEM-Derived Morphology Quality Index (MQI). YOLO-derived grain-center priors guide U-MambaBot segmentation, while Class-Probability TV Regularization improves local coherence. Six descriptors are extracted from the segmented ${\mathrm{ABX}}_3$ matrix and ${\mathrm{PbI}}_2$ -like regions, from which four are selected for MQI construction. Across nested, publication-grouped, and within-publication validation schemes, MQI retains a significant monotonic association with power conversion efficiency (Spearman $\rho =0.34$ and 0.42). In annealing-time and additive-concentration studies, MQI reproduces film- and device-level performance trends and identifies the same optimal processing conditions. Without refitting, it further tracks reported morphology-performance trends across 12 independent studies and 37 processing conditions, yielding a pooled within-study Spearman $\rho =0.90$ (publication-level bootstrap 95% CI 0.79-1.00). This framework enables quantitative, morphology-aware SEM screening and process-condition ranking for perovskite films.