LSTM-Driven Impact Burial Prediction with Self-Attention and SE Model Integration

To address the limitations of current research on impact burial prediction, we propose a novel hybrid neural network model, LSTM-SA-SE, which integrates a Long Short-Term Memory (LSTM) network, a Self-Attention (SA) mechanism, and a Squeeze and Excitation (SE) module. The architecture is composed of five components: an input layer, LSTM layer, SA layer, SE layer, and output layer. In this framework, the LSTM captures temporal dependencies, the SA mechanism models internal feature relationships, and the SE module enhances the network's dynamic feature recalibration capabilities. Compared with conventional deep learning models such as LSTM, LSTM-SA, and LSTM-SE, the proposed model demonstrates enhanced effectiveness in capturing complex spatiotemporal patterns and adaptively refining features. This leads to more accurate predictions of impact burial behavior, significantly enhancing overall estimation accuracy.