Analysis of Graphic Images to Expand Descriptions of Physical Effects

Often, the description of the physical effect in the “primary” sources of information (patent documents, journal articles of a physical profile, dissertations in physics) contains graphs of dependencies linking the physical quantities of the input and output impacts of the physical effect (PhE). The analysis of this information and its use to expand the description of the physical effect is an urgent task. A modification of the model for describing the physical effect was carried out, taking into account the graphical representation of the functional dependence of the physical quantities of the PhE. CNN, LSTM, Visual Transformer (ViT) neural networks and machine learning methods of Naive Bayes, SVM have been tested in relation to classification and recognition tasks on images of dependency graphs. A comparative analysis of the approaches (accuracy of definition, complexity of data preparation, fault tolerance, equipment requirements) was carried out using criteria of precision, recall, F-measure, AUC-ROC.