Permanent magnet synchronous motor (PMSM) is subjected to frequent electric-magnetic-force-thermal shocks for a long period of time, which accelerates the aging of winding insulation and leads to the occurrence of high-resistance connection (HRC) faults. HRC faults further induce more serious damages to the PMSM, making their accurate diagnosis highly significant. At present, based on the evolution law of PMSM operating voltage and load current, it can provide reference for accurate identification of HRC. However, the above are invasive methods, potentially causing interference with the normal operation of the motor. Since HRC faults significantly alter the electromagnetic field distribution within the motor, the leakage signals in the motor space can serve as an alternative, non-invasive source of state information. By acquiring these leakage signals, a non-invasive approach to diagnosing HRC faults can be realized. To this end, an intelligent diagnosis method for PMSM high-resistance faults based on electromagnetic multidimensional spatio-temporal characteristics is proposed. This method establishes the correlation relationship between spatial leakage signals and the motor′s differentiated state, enabling the intelligent motor state evaluation by a joint intelligent algorithm. First, the electromagnetic signal evolution law under fault is analyzed based on the equivalent circuit model of the motor winding, identifying the optimal electromagnetic test point. Secondly, the feature image conversion and dimensioning method based on the array of leakage signals are proposed to diagnose motor faults with GoogLeNet network. Finally, the proposed method is verified by simulation model and experimental platform. The experimental results show that the feature image upscaling and intelligent assessment method based on the leakage signal array can accurately identify and locate HRC while also assessing fault severity, achieving an accuracy rate of up to 97%, which verifies the effectiveness of the proposed method. The method has the advantages of non-invasive and high accuracy, and has a wider application prospect for PMSM.