•  

Development of a heterogeneous object recognition model based on nested machine learning methods

TitleDevelopment of a heterogeneous object recognition model based on nested machine learning methods
AuthorsK. N. Novikova1, S. V. Novikova1
1Kazan National Research Technical University named after A. N. Tupolev
AnnotationThe article proposes a method for identifying objects in the case when their type (nature, origin) is not known in advance. In existing systems of identification, or recognition, as a rule, the origin, or general type, of objects is known in advance. For instance, encrypted digital cardiograms are fed into the system, and the system must detect pathology. Either fingerprint images are fed into the system, and the task of the model is to determine their owners. The situation when the type of the identified object is not known is practically never encountered. This paper proposes a two-level intellectual model that allows at the first stage of its operation to identify the type of object (for example, whether the object is an image, an encrypted cardiogram, telemetry readings, etc.), and at the second stage to determine its type directly (if the object is a cardiogram, it is necessary to determine pathology). The first level includes a clustering model based on the k-means algorithm. The second level forms an ensemble of CNN type models. The structure of convolutional networks is determined individually for each cluster (object type) selected at the previous level.
Keywordsheterogeneous objects, object identification, object clustering, machine learning models, nested models.
CitationNovikova K. N., Novikova S. V. ''Development of a heterogeneous object recognition model based on nested machine learning methods'' [Electronic resource]. Proceedings of the International Scientific Youth School-Seminar "Mathematical Modeling, Numerical Methods and Software complexes" named after E.V. Voskresensky (Saransk, July 14-18, 2022). Saransk: SVMO Publ, 2022. - pp. 149-159. Available at: https://conf.svmo.ru/files/2022/papers/paper25.pdf. - Date of access: 29.03.2024.