One　critical　challenge　to　achieving　reliable　wind　turbine　blade　structural　health　monitoring　(SHM)　is　mainly　caused　by　composite　laminates　with　an　anisotropy　nature　and　a　hard-to-access　property.　The　typical　pitch-catch　PZTs　approach　generally　detects　structural　damage　with　both　measured　and　baseline　signals.　However,　the　accuracy　of　imaging　or　tomography　by　delay-and-sum　approaches　based　on　these　signals　requires　improvement　in　practice.　Via　the　model　of　Lamb　wave　propagation　and　the　establishment　of　a　dictionary　that　corresponds　to　scatters,　a　robust　sparse　reconstruction　approach　for　structural　health　monitoring　comes　into　view　for　its　promising　performance.　This　paper　proposes　a　neighbor　dictionary　that　identifies　the　first　crack　location　through　sparse　reconstruction　and　then　presents　a　growth　sparse　pursuit　algorithm　that　can　precisely　pursue　the　extension　of　the　crack.　An　experiment　with　the　goal　of　diagnosing　a　composite　wind　turbine　blade　with　an　artificial　crack　is　performed,　and　it　validates　the　proposed　approach.　The　results　give　competitively　accurate　crack　detection　with　the　correct　locations　and　extension　length.