Segmented Matching Method of Multi-Geophysics Field SLAM Data Based on LSTM

At present, simultaneous localization and mapping (SLAM) has become an important method for autonomous underwater vehicles (AUVs) to realize long-term navigation. However, using only bathymetric data in unknown environment has its own disadvantages, that are low precision and large computational load. To tackle with requirements of high-precision navigation under large-scale and long-term voyage condition, a SLAM method and corresponding matching algorithm for integrating multi-geophysical field data are proposed. By dividing the feature data and location data of geophysical field obtained into various submaps and sub-segments during AUV sailing, the dominant navigation data of each segment is identified using long short-term memory network. Validity of the proposed method is done by simulation experiments. During the simulation, the loop closure detection of each submap is used, and the matching counter is set to check the correct matching rate. Finally, the matching results with single geophysics field data under the same conditions are compared with multi-geophysics field data and analyzed. The experimental results have demonstrated the feasibility and correctness of the proposed method.