A Fast Classification Method for Ground Motion Targets Based on Seismic Perception Node and Attention Mechanism

The most urgent task in the seismic perception system is to realize the efficient classification of ground-moving targets. This letter uses embedded adaptive learning technology to solve the timeliness problem of real-time moving target monitoring. For the target recognition task, a neural network based on an attention mechanism was proposed, which effectively paid attention to the temporal features and the deep nonlinear features. The spatial attention mechanism is added to concentrate on the weight of useful information, which improved the accuracy of target recognition and effectively saved time. The performance of the proposed method is evaluated by different target signals collected in this deployment. Whether in a normal dataset or a noisy dataset, the dense neural network model based on an attention mechanism achieves an accuracy rate of more than 97% in the classification tasks of five types of targets, and at the same time greatly improves the computational efficiency. This solves the problems of time-consuming feature extraction and insufficient robustness.