Research on real-time compensation method of UAV aeromagnetic based on deep learning

——The use of neural networks in aeromagnetic compensation is effective and popular, and compared with least squares method, the use of neural networks has higher requirements for computing power. However, the computing resources on the unmanned aerial vehicle (UAV) flight detection platform are limited, so the existing neural network needs to be optimized to achieve the purpose of real-time computing under the limited computing resources. In the process of conducting this experiment, it is found that the training time of BP neural network is relatively large, therefore, we will use BP neural network pre-training for its weight, and then use transfer learning to the training results, which will greatly reduce the operation time and improve the operation efficiency. In addition to this, we have applied other methods to accelerate the neural network to achieve rapid training of the BP neural network on the Raspberry Pi system.Index terms—low-rank decomposition, parameter quantization, real-time aeromagnetic compensation, transfer learning, unmanned aerial vehicles mobile detection platform.