Taking　advantages　of　both　k-mean　clustering　algorithm　and　mathematical　morphology,　an　image　processing　method　was　proposed　for　ultrasonic　phased　array　images　to　improve　the　image　feature　extraction　and　then　to　facilitate　automatic　defect　detection　of　Polyethylene　(PE)　pipe　electrofusion　joints.　K-mean　clustering　algorithm　is　powerful　in　image　segmentation　whilst　mathematical　morphology　has　superiority　in　image　edge　smoothing　and　local　image　segmentation.　The　results　demonstrate　that　the　proposed　method　does　not　need　the　training　samples　and　it　does　self-training　using　the　properties　of　a　single　image.　It　can　extract　multiple　defect　information　from　ultrasonic　phased　array　images　effectively,　which　provides　a　new　technique　for　nondestructive　testing　and　evaluation　of　defected　PE　pipe　electrofusion　joints　and　it　can　be　recommended　for　engineering　applications.　©　2018,　Editorial　Board　of　Transactions　of　the　China　Welding　Institution,　Magazine　Agency　Welding.　All　right　reserved.