Geospatial indexing for sea–land navigation based on machine learning

The exponential growth of geospatial data has created a need for efficient data processing and indexing, especially for acquiring the required data around while traveling on land and sea. Machine learning based learned index has brought a new phase in the field of spatial data indexing on land and sea for fast and accurate acquisition of points of interest. Recently proposed spatial learned indexes and traditional spatial indexes are difficult to maintain their performance while traveling on land and sea due to high memory cost and time consumption. Considering the above issue, we propose SLIP, a novel in-memory spatial learned index. SLIP uses an optimized mapping function and a piecewise linear model to handle data updates (e.g., insertion and deletion) efficiently. Based on SLIP, we design a range query algorithm and a faster KNN(K-NearestNeighbor) query algorithm using the optimized radius expansion model to get the results more quickly. Experiments on real geospatial datasets and synthetic datasets used the electronic navigation chart software show that SLIP achieved an average improvement of 10x in query performance over traditional spatial indexes and state-of-the-art learned indexes. Therefore, SLIP provides an innovative idea for high-efficiency geospatial data processing and indexing, and offers an efficient method for acquiring the surrounding geographic data for future navigation on land and sea.