Intelligent　tunnelling　has　become　an　important　direction　for　the　development　of　TBM　technology　recently.　As　a　result　of　the　interaction　between　rock　mass　and　TBM　cutterhead,　mucks　are　very　important　for　predicting　rock　mass　conditions　and　evaluating　rock　breaking　efficiency.　A　real-time　muck　analysis　system　for　assistant　intelligence　TBM　tunnelling　is　proposed　in　this　paper.　Machine　vision　was　applied　to　take　the　muck　images　continuously　in　the　high-speed　conveyor　belt.　The　image　segmentation　and　feature　extraction　of　the　mucks　are　conducted　by　using　a　deep　learning　algorithm.　The　proposed　system　also　measured　the　mass　and　volume　flow　of　the　muck　by　installing　a　belt　scale　and　a　scanner　to　monitor　the　stability　of　the　rock　mass　on　the　tunnel　face.　After　the　system　was　completed,　it　was　installed　on　an　indoor　simulation　experimental　platform.　A　series　of　experiments　were　conducted　to　verify　the　design　functions　and　measurement　accuracy.　Additionally,　the　system　was　applied　to　a　TBM　tunnelling　project.　The　application　results　showed　that　the　proposed　system　reached　its　design　requirements　and　functions,　and　can　provide　muck　data　support　for　further　assistant　intelligent　TBM　tunnelling.