Abstract: Rockbursts and stress-induced collapses constitute predominant high-stress hazards in deep-buried tunnels.Owing to their analogous failure processes and controlling factors,accurate differentiation between these two disaster types remains challenging.Nevertheless,significant disparities exist in the block size distribution characteristics of failed rock masses.This study investigates a representative stress-induced disaster case in a deep-buried tunnel,developing (i) an image filtering model integrating homomorphic and bilateral filtering techniques,and (ii) an intelligent recognition system combining binarization with the Watershed Algorithm for block size analysis.The proposed methodology enables machine vision-based identification of failed masses and quantitative characterization of block size distributions in both rockburst and stress-induced collapse events.Microseismic monitoring data are incorporated to elucidate the mechanistic origins of block size variations.Key findings include:(1)While most rock blocks in stress-induced failures can be effectively segmented,challenges persist in distinguishing thinly layered and clastic debris.(2)The “04.21” moderate rockburst exhibited the highest block count,with an overall larger failure block size,whereas the “08.13” stress-induced collapse demonstrated minimal block quantities and finer fragmentation.The failure block size quantity of the “05.04” moderate rockburst was intermediate between the two other events.(3)Clastic debris predominates in stress-induced collapses,with rockburst intensity positively correlating with both mean fragment size and large-block proportion.(4)Microseismic energy release and event frequency exhibit positive correlations with fragment size in stress-induced disasters,while shear failure mechanisms govern clastic debris formation in collapse events.This research establishes a novel quantitative-intelligent framework for classifying tunnel stress disasters,concurrently advancing mechanistic analysis through fragmentation characterization.