Abstract: In view of problem of difficulties in obtaining prior knowledge during fusion of heterogeneous data in mining field, poor real-time performance of IoT ontology database,and low efficiency of manual annotation of instance object data, an automatic semantic annotation method for mine Semantic Web of things was proposed. Framework of semantic processing of sensory data was given: on the one hand, professional domain and category of ontology are determined, and the domain ontology is constructed by reusing SAO as the basis for driving semantic annotation; on the other hand, machine learning method is used for feature extraction and data analysis of perceptual data stream, and relationship between concepts is mined from massive data; finally, data mining knowledge is used to drive the update and improvement of the ontology, so as to realize dynamic update, expansion and more accurate semantic annotation of the ontology, and enhance the machine's understanding. Spindle fault of mine hoisting system is used as an example to explain the process of semantic annotation from ontology to instantiation: combining the domain expert's knowledge and ontology reuse, the “seven-step method” is used to establish fault ontology of the main drive of mine hoisting system; in order to enhance the accuracy of the instance data attribute description, PCA dimensionality reduction method and K-means clustering method are used to group the data set to extract the relationship between data attributes and concepts;finally, the relationship between specific preconditions and subsequent concepts is marked by SWRL to optimize the domain ontology. The experimental results show that in the process of ontology instantiation, machine learning technology can be used to automatically extract concepts from sensing data and realize automatic semantic annotation of sensing data.