Particle　Filter　(PF)　is　well　known　as　a　state　estimation　method　for　nonlinear　and　non-Gaussian　system.　However,　PF　has　the　inherent　drawbacks　of　loss　of　diversity　in　particles　due　to　sampling-importance　resampling　(SIR).　The　loss　may　lead　to　PF＇s　worse　performance,　even　making　filtering　diffuse.　In　this　paper,　a　Linear　Optimization　Particle　Filter　(LOPF)　based　on　linear　optimization　resample　was　proposed　to　overcome　the　drawbacks　of　the　loss　of　diversity　in　particles.　Based　on　the　linear　optimization　resample,　the　new　algorithm　calculates　the　new　particles　in　the　next　deviation　according　to　the　linear　combination　of　the　abandoned　particles　and　copied　particles.　These　particles　were　used　to　replace　the　copied　particles　by　SIR　to　overcome　the　loss　of　diversity　caused　by　resampling　only　from　the　particles　with　large　weight.　Then　LOPF　was　applied　to　nonlinear　initial　alignment　for　the　strapdown　inertial　navigation　system　with　large　azimuth.　The　simulation　results　demonstrate　that　LOPF　is　better　than　PF　in　the　alignment　time　and　alignment　accuracy.　Therefore　LOPF　is　more　suitable　to　the　strapdown　inertial　navigation　system＇s　nonlinear　initial　alignment.