Since　more　and　more　serious　low-frequency　oscillation　phenomena　have　happened　in　interconnected　power　grids,　a　high-accuracy　low-frequency　oscillation　identification　method　is　proposed　to　overcome　the　shortages　of　the　existing　methods.　The　method　is　based　on　the　opening　and　closing　operations　of　generalized　morphology　to　design　an　improved　generalized　morphological　filter,　which　can　effectively　eliminate　the　noise　and　retain　the　original　features　of　signals.　An　advanced　matrix　pencil　algorithm　was　proposed　to　identify　parameters　from　low　frequency　oscillation　signals.　A　standardized　singular　entropy　technique　was　utilized　to　solve　the　key　problem　of　order　determination.　By　this　way　the　estimating　value　of　the　order　can　be　very　close　to　the　real　value　in　the　power　system,　which　enhances　identification　accuracy.　Simulations　verified　the　proposed　low-frequency　oscillation　identification　method.　©　2017,　The　editorial　office　of　Transaction　of　China　Electrotechnical　Society.　All　right　reserved.