A Study On The Construction Of Knowledge Graph In The Field Of Remote Sensing Based On Machine Learning And Fuzzy Decision

The application research in the field of remote sensing is usually highly related to sensors.At the same time,for detecting targets,a single sensor can no longer meet the needs.The collaborative work of multiple sensors is the general trend.This makes researchers in the field of remote sensing focus on “what sensors are used in what applications”.However,there are a large number of sensors and applications in related fields,and it is difficult for researchers and applications personnel to quickly grasp the state of the industry.Therefore,research on the automatic processing of related knowledge text and the construction of knowledge graphs based on the field of remote sensing “what sensors are used in what applications” has important value.This paper focuses on this problem and studies the construction of knowledge graphs of knowledge pairs such as “Sensors”-“Applications” in the field of remote sensing.The research process of this method can be extended to the construction of knowledge graphs of other aspects of the field of remote sensing.The specific work is divided into the following three parts:(1)The method of short text feature extraction and classification in the field of remote sensing based on machine learning is studied.Traditional text mining methods cannot be directly applied to scientific text mining tasks.In order to solve this problem,the goal is to judge whether the sentences of remote sensing documents are “Sensors” or “Application”.By fusing a variety of different feature extraction methods to find the best classification features,three machine learning classification algorithms are used for verification.For the“Sensors” classification task,the precision,recall and F-measure reached 86.6%,91.0% and88.7% respectively;for the “Applications” classification task,the precision,recall and F-measure reached 86.9%,80.3% and 83.4% respectively.Therefore,the feature extraction and classification methods used can automatically identify “Sensors” and “Applications”.(2)A text classification algorithm based on fuzzy decision is established to further improve the accuracy of a single classifier.Through the knowledge of intuitionistic fuzzy sets in fuzzy decision theory and multi-attribute group decision-making,and the classification result of the three classifiers in the part of(1),the intuitionistic fuzzy preference relationship matrix is determined to obtain the standard priority of the classifier weights.Finally,a decision matrix is constructed according to the sample membership to determine the classification result.For the “Sensors” classification task,the precision,recall and F-measure are 90.0%,95.2% and 92.5%,respectively,and the F-measure increased by3.8% compared with the part of(1);for the “Applications” classification task,the precision,recall and F-measure are 85.2%,84.3% and 84.7%,respectively,and the F-measure is increased by 1.2% compared with the part of(1).The classification results show that the algorithm can improve the classification performance of a single classifier and the performance of automatic recognition of “Sensors” and “Applications” in the field of remote sensing.(3)The method of constructing the knowledge graph of the “Sensors”-“Application”knowledge pair in the field of remote sensing is studied.Through the named entity recognition of the positive samples in the first part and the second part,automatically extract the triples {“RS”: Sensors;“ACT”: Roles;“APL”:Applications},and construct the knowledge graph.In order to solve the problem of a small number of training samples and weak classification performance,the transfer learning is used.In the end,the “RS” entity recognition F-measure is 77.0%,the “ACT” entity recognition F-measure is 68.8%,and the“APL” entity recognition F-measure is 47.5%.The entities are visualized the knowledge graph through the Neo4 j graph database.This paper proposes an automatic recognition method for knowledge pairs such as“Sensors”-“Applications” in the remote sensing literature.The knowledge graph automatically constructed by scientific text can help people in the industry know which applications a certain sensor has and which sensors can be associated with a certain application.The graph can quickly grasp the state of the industry and save a lot of time.