Research On The Construction Of Knowledge Annotation Corpus For Disaster Emergency Remote Sensing Image Application Cases

Task-oriented remote sensing image data retrieval for disaster response is one of the key methods to improve emergency response capabilities.knowledge of disaster emergency remote sensing image application cases is the foundation for achieving remote sensing image data retrieval driven by disaster emergency remote sensing tasks.The academic text summary of disaster emergency remote sensing application contains a large number of disaster emergency remote sensing image application case knowledge.However,the academic text summary is a kind of unstructured data,which needs to use Natural language processing and other technologies to extract knowledge.Text corpora are important data resources for Natural language processing model training and learning,which annotate named entities and relationships in the text.Therefore,building a annotated corpus is of great significance for acquiring disaster emergency remote sensing image application case knowledge.This article constructs a feature model for disaster emergency remote sensing application cases based on the application of remote sensing in disaster emergency response;By analyzing the characteristics of academic text abstracts for disaster emergency remote sensing applications,an academic text annotation system was established,and a knowledge annotation standard for disaster emergency remote sensing image application cases was designed;A corpus was annotated using academic text abstracts as the data source.This work mainly consists of three parts:(1)Design of Academic Text Annotation System for Remote Sensing Image ApplicationsBased on the application of remote sensing in disaster emergency,this paper analyzes the composition and relationship of elements in disaster emergency remote sensing application cases,and establishes a feature model for disaster emergency remote sensing application cases;A remote sensing image application academic text annotation system was designed by analyzing the characteristics of academic text abstracts for disaster emergency remote sensing applications.(2)Specification for knowledge labeling of disaster emergency remote sensing image application casesThe named entities and the relationships between entities in the academic text abstract of disaster emergency remote sensing applications are complex and diverse.In order to standardize corpus annotation,entity annotation specifications have been designed for disaster emergency remote sensing task entities,remote sensing data entities,method entities,effect entities,time entities,and spatial entities;For the relationship between entities,we have designed sentence level and discourse level relationship annotation standards.(3)Constructing Knowledge Annotation Corpus of Remote Sensing Image Application CasesThere is no publicly annotated corpus in the field of disaster emergency remote sensing image application.Based on the knowledge annotation specification designed in this article,annotate academic text abstracts;During the annotation process,a deep learning model is used to validate the corpus and optimize the annotation process;Finally,a corpus consisting of 2000 academic text abstracts was constructed,which included a total of 31575 entities and 15424 relationships between entities.