Tunneling magnetoresistance (TMR) sensors have shown the capability of operating in weak magnetic fields. However, the environmental magnetic noise limits their applications in open field detection. This article proposes a novel background noise cancellation method based on a backpropagation (BP) neural network for TMR sensor arrays. According to simulation results, the BP-based noise reduction method can eliminate background noise more effectively than the traditional coherence coefficient method. The signal-to-noise ratio (SNR) of the sensor can, thus, be improved by over 20 dB, especially when detecting extremely low SNR signals. This algorithm is demonstrated using a TMR sensor array, which shows a capability of greatly enhancing the sensor array’s limit of detection in open field testing.