Magnetic induction tomography (MIT) is a promising technique for cerebral hemorrhage detection due to its non-radioactive, non-invasive, low cost features and capacity to penetrate the low-conductivity skull. In the cerebral hemorrhage detection by MIT, multi-frequency decomposition method can reduce the artifacts caused by the background conductivity changes in the existing frequency-difference method. In order to improve the accuracy of multi-frequency decomposition, a cambered sensor array with focusing coil was proposed. The performance of focusing coil was tested by 3D anatomical head models with different hemorrhage depths. Target to noise ratio was investigated to evaluate the antiinterference performance. Decomposition error and correlation coefficient were calculated to evaluate the accuracy of multifrequency decomposition. After introducing the focusing coil, the decomposition error is reduced by 39.0%, 25.6% and 15.3% respectively at the hemorrhagic depth 12 mm, 16 mm and 20 mm. The results show that the focusing coil is effective in enhancing the anti-interference performance against the noise signal generated by the conductivity changes of other tissues and improving the accuracy of multi-frequency decomposition. This method provides the possibility of cerebral hemorrhage detection with smaller volume in frequency-difference MIT.