The abstract should summarize the contents of the paper in short terms, i.e. 150–250 words. When the low-field magnetic resonance equipment operates in an unshielded environment, EMI noise will be received by the radio frequency coil during the acquisition of k-space data, resulting in issues reduced signal-to-noise ratio, thereby affecting the accuracy of clinical diagnosis. Focusing on this problem, this paper adopts EMI detection coils to collect EMI noise in the equipment environment, forming input output data pairs with the radio frequency coil, and mapping the noise influence within the main coil through neural networks. The distribution of EMI coils and the minimum data set are optimized through correlation analysis and orthogonal experimental design. Considering the differences in signal amplitude characteristics of different EMI sampling coils, clustering analysis and branch convolution mechanisms are introduced on the basis of the convolutional neural network to extract noise signal features from sampling coils at different positions. Consequently, while using the minimum amount of data from sampling coils, the noise mean square error is reduced by 13%, enhancing the data utilization efficiency.