Decarburization　refers　to　a　phenomenon　that　the　content　of　carbon　in　metal　is　reduced,　and　this　could　be　advantageous　or　detrimental　depending　on　the　usage　scenario.　In　this　paper,　a　new　feature,　called　Barkhausen　frequency　scale　cepstrum　coefficients　(BFCCs),　which　was　based　on　the　cepstrum　was　proposed　to　evaluate　the　depth　of　the　decarburized　layer.　The　Genetic　algorithm　was　used　to　optimize　the　parameters　of　the　BFCCs　feature.　In　the　experiment　part,　the　regression　precision　was　compared　under　BFCCs　and　MBN　envelope　features　using　multiple　linear　regression　(MLR)　and　BP　neural　network　(BP).　The　regression　experiment　result　was　measured　by　mean　absolute　error　(MAE)　and　shows　the　good　representative　ability　of　BFCCs　for　the　depth　of　the　decarburized　layer.　©　Published　under　licence　by　IOP　Publishing　Ltd.