A　novel　detection　method　and　targets　signal　extraction　algorithm　for　parallel　steel　pipeline　localization　and　defects　recognition　based　on　harmonic　magnetic　field　are　proposed.　An　orthogonal　vector　differential　focusing　array　detector　was　designed,　which　radiates　high-frequency　(HF)　and　ultra-low-frequency　(ULF)　electromagnetic　waves　to　pipeline　based　on　the　principle　of　Frequency　Modulated　Continuous　Wave　(FMCW),　and　records　the　signals　in　parallel　using　high-sensitivity　Tunnel　Magneto　Resistance　(TMR)　array　sensors.　The　joint　signal　preprocessing　algorithm　of　Combination　Morphological　Pulse　Elimination　(CMPE)　and　Quantum　Genetic　Gaussian　Potential　Stochastic　Resonance　(QGGPSR)　is　proposed　to　improve　signal　quality　by　adaptively　eliminating　random　pulse　and　noise.　The　time-frequency　analysis　algorithm　Complementary　Ensemble　Local　Mean　Decomposition　(CELMD)　is　used　to　decompose　the　pipeline　defects　signal　in　detail,　and　the　appropriate　components　are　selected　and　fused　into　a　correlation　array　signal　cluster.　Finally,　the　targets　signal　are　separated　by　Limited-memory　BFGS　Independent　Component　Analysis　(L-BFGS-ICA)　based　on　machine　learning　optimization　to　enhance　their　legibility.　The　magnetic　dipole　harmonic　simulation　model　and　the　20#　steel　parallel　pipeline　experimental　platform　are　established,　and　compared　with　the　passive　detection　method　to　verify　the　effectiveness　and　practicality　of　the　proposed　method.　The　results　show　that　this　method　can　accurately　locate　parallel　pipelines　and　detect　different　kinds　of　defects　in　noisy　environment,　which　is　of　great　significance　for　the　application　of　non-contact　Non-Destructive　Testing　(NDT)　in　practical　engineering.