Lightweight fuzzy semantic precise characterization and security adaptation method

Abstract

The present application relates to the field of rail transit automatic driving technology, and discloses a light-weight fuzzy semantic accurate description and safety adaptation method, comprising: receiving continuous frames of track data and multi-source original semantic data, performing a dynamic pruning operation on a neural network model in a light-weight semantic redundancy filtering module through a parameter dynamic pruning module, obtaining a light-weight semantic redundancy filtering module, and obtaining simplified semantic data; inputting the simplified semantic data into a fuzzy semantic quantification module to obtain a quantitative evaluation of fuzzy semantics; when the risk value of the quantitative evaluation exceeds a first preset threshold, initiating a cooperation request with a track inspection robot through a risk traceability interface based on a standard communication protocol, and obtaining an on-site monitoring review result fed back by the track inspection robot; fusing and determining based on the quantitative evaluation value and the review result, and generating and issuing a safety control instruction according to the determination result, so that the fuzzy semantic capture rate can be improved and the computing power consumption can be reduced.

Images