Civil engineering supervision video retrieval method optimization based on spectral clustering and R-tree

摘要

The civil engineering supervision video provides the effective method to improve the quality of civil engineering supervision, but its usual retrieval by B+ tree can't show the efficient performance to meet the real requirements. This paper uses some natural language processing ways, such as word embedding and combines semantic, to let the machine realize the content of supervision video and then focuses on the civil engineering supervision video retrieval annotated by supervision engineer. Firstly, we described the civil engineering supervision video hierarchical model with semantic, its framework and storage. And we proposed a CESVSR-tree data process algorithm to transform the civil engineering supervision video annotation into word vector through Chinese Wikipedia Entries and civil engineering entries, get the word weight value of each word. Secondly further research on video data index, we proposed the spectral clustering-based node split algorithm, it combines the traditional R-tree node splitting algorithm with spectral clustering algorithm, which improves the indexing speed of high-dimensional data such as video and word vector. Finally, in view of the rapid development of solid-state driver, this paper optimized the R-tree with the characteristics of solid-state driver, to improve the index construction speed on the hybrid storage structure.