Doubly salient electromagnetic machine (DSEM) can constitute a competitive rugged starter generator (SG) system. But under the traditional startup control method, the field current is set to the rated value regardless of speed or load torque conditions, and the phase currents are controlled with standard angle commutation, bringing in large power loss as well as non‐negligible torque ripple. To overcome these shortcomings, based on the established DSEM power loss calculation model, a currents distribution strategy is proposed according to the obtained quantitative relationship between DSEM power loss and field current under multiple speed and load torque conditions. Then, to identify the load torque for practical implementation, a novel torque observer is designed based on back propagation (BP) neural network algorithm. Afterwards, to further improve the startup torque performance, currents distribution with advanced angle commutation (AAC) strategy is put forward, where the optimal advanced angle is selected from a 3‐D lookup table to achieve the minimal commutation torque ripple. Under the proposed control strategies, DSEM power loss as well as the output torque ripple are desired to be decreased, the simulation and experimental results on a 12/8‐pole DSEM prototype verify the correctness and feasibility of the proposed strategies under multiple operating conditions.